Comparing changes

Choose two branches to see what’s changed or to start a new pull request. If you need to, you can also .

Open a pull request

Create a new pull request by comparing changes across two branches. If you need to, you can also .

Checking mergeability… Don’t worry, you can still create the pull request.

36 contributors

This comparison is big! We’re only showing the most recent 250 commits

Commits on Oct 03, 2018

tschottdorf


      log: include a call to action with fatal errors

When rare errors happen in the wild, they are a) often unreported or b)
reported only in anonymized form, with little context that can help
pinpoint the root cause.

Users can help us out tremendously by contacting us, and so we should
incentivize that.

Do so by concluding fatal errors (i.e. most crashes) with a call to
action.

Touches #28699.
Touches #24033.
Touches #25173.

Release note: None

a518de5

couchand and tschottdorf


      log: minor copyedits to fatal error postamble

Release note: None

7f1da0b

nvanbenschoten


      Merge pull request #30920 from nvanbenschoten/backport2.1-30909

release-2.1: storage: grab RLock in Replica.propose when checking destroyStatus

71dd3cb

a-robinson


      Merge pull request #30921 from a-robinson/backport2.1-30907

release-2.1: storage: Avoid printing writes-per-second info to logs and rangelog

e65a612

m-schneider


      core: fix and add test for per locality hosts

Connected the existing code and added an acceptance roachtest.

closes: 29591

Release note: None

a7b24d6

m-schneider


      release-2.1: core: fix and add test for per locality hosts

144d5a5

mjibson


      security: use same error message for user DNE and bad password

This prevents attacks to determine which users exist.

Fixes #30879

Release note (bug fix): "user does not exist" and "invalid password"
errors now produce the same error message during password login.

ad5bf3e

danhhz


      Merge pull request #30903 from danhhz/backport2.1-30896

release-2.1: changefeedccl: fix panic when pk is updated and a column is not nullable

6f93127

ridwanmsharif


      sql: remove depreceted config_yaml field from output

Since `cockroach zone` is a deprecated command and we
no longer support the YAML method to configure zones, we shouldn't
show the YAML in the output at all for `SHOW ALL ZONE CONFIGURATIONS`
and for `SHOW ZONE CONFIGURATION FOR ...`. For backwards compatibility,
we can not remove the column all together yet so instead make the column
hidden.

Release note: None

44e891f

Commits on Oct 04, 2018

knz


      cli/sql: avoid a panic upon initial session errors

Prior to this patch, if the server closed the connection (because of
some internal error) in-between the initial pq.Open() and the middle
of conn.getServerMetaData(), the client conn object would become
invalid. However this error case was turned into a warning and
subsequent accesses by the client code would then in turn into a
panic (instead of a clean "connection error").

This patch fixes this by ensuring the connection is properly cleaned
up if the intermediate SQL query in getServerMetadata() fails.

Release note: None

knz


      cli/start,server: fix the "wait for init" msg

Release note (bug fix): Previously `cockroach start` would erroneously
hint that it was waiting for a join or `cockroach init` both for fresh
new nodes and re-starting nodes. This has been fixed and the message
now only appears for fresh new nodes.

abc3169

knz


      Merge pull request #30952 from knz/backport2.1-30688

release-2.1: cli/sql: avoid a panic upon initial session errors

26b405f

knz


      Merge pull request #30953 from knz/backport2.1-30211

release-2.1: cli/start,server: fix the "wait for init" msg

219c0ba

vilterp


      Merge pull request #30892 from vilterp/backport2.1-30842

release-2.1: ui: add semicolon to CREATE USER command on login screen

bde7afc

craig[bot] and ridwanmsharif


      Merge #30937

30937: release-2.1: sql: remove depreceted config_yaml field from output r=ridwanmsharif a=ridwanmsharif

Backport 1/1 commits from #30927.

/cc @cockroachdb/release

---

Since `cockroach zone` is a deprecated command and we
no longer support the YAML method to configure zones, we shouldn't
show the YAML in the output at all for `SHOW ALL ZONE CONFIGURATIONS`
and for `SHOW ZONE CONFIGURATION FOR ...`. For backwards compatibility,
we can not remove the column all together yet so instead make the column
hidden.

Release note: None


Co-authored-by: Ridwan Sharif <ridwan@cockroachlabs.com>

a167168

RaduBerinde


      opt: fix crash caused by tuple IN tuples

Some of the constraint generation code relies on the values in the
right-hand side tuple of an IN expression being sorted (which is done
by NormalizeInConst). However, this rule does not sort tuples of
tuples.

Fixing the code to call `spans.MergeAndSort`. Note that this function
has a fast-path if the spans are already sorted.

Release note (bug fix): Fixed a possible crash when using filters with
"tuple IN tuples" expressions.

a-robinson


      storage: Add minimum cluster version for load-based rebalancing

This is needed because v2.0 didn't gossip QPS data, so in a
mixed-version cluster the 2.1 nodes will think all the 2.0 nodes have no
load on them and continually try to give them more leases and replicas.

Gossiping QPS data was added in #27857.

Release note: None

942a80d

a-robinson


      Merge pull request #30971 from a-robinson/backport2.1-30964

release-2.1: storage: Add minimum cluster version for load-based rebalancing

ee7af94

RaduBerinde


      Merge pull request #30968 from RaduBerinde/backport2.1-30958

release-2.1: opt: fix crash caused by tuple IN tuples

f895b99

a-robinson


      partitionccl: Clarify license check error when setting zone configs

Fixes #30960 in the laziest way possible. We could presumably also scan
through the subzones to determine whether they're for indexes,
partitions, or both, but this seems clear enough for people to
understand.

Release note: None

a-robinson


      storage: Don't transfer leases to behind replicas in StoreRebalancer

Transferring a lease to a replica that's behind can cause all requests
to the range to stall, as the old leaseholder thinks it's no longer the
leaseholder but the new leaseholder doesn't know it's the leaseholder
yet. This avoids creating such scenarios in the StoreRebalancer.

Release note (bug fix): Avoids an edge case in load-based
rebalancing where we could transfer the lease for a range to a replica
that isn't keeping up with the other replicas, causing brief periods
where no replicas think they're leaseholder for the range and thus no
requests can be processed for the range.

24b7e2c

a-robinson


      Merge pull request #30978 from a-robinson/backport2.1-30962

release-2.1: partitionccl: Clarify license check error when setting zone configs

93a54f9

Commits on Oct 05, 2018

tschottdorf


      Merge pull request #30922 from tschottdorf/backport2.1-30822-30898-30918

backport-2.1: log: include a call to action with fatal errors

e29115d

knz


      sql: improve the table name prefix resolution error message

Before:

```
root@127.0.0.1:44824/defaultdb> create table t.t(x int);
pq: no schema has been selected to create "t.t" in
HINT: verify that the current database and search_path are valid
```

After:

```
root@127.0.0.1:36392/defaultdb> create table t.t(x int);
pq: cannot create "t.t" because the target database or schema does not exist
HINT: verify that the current database and search_path are valid and/or the target database exists
```

Release note: None

528951e

knz


      Merge pull request #30454 from knz/backport2.1-29623

release-2.1: sql: improve the table name prefix resolution error message

d1504f6

a-robinson


      Merge pull request #30972 from a-robinson/backport2.1-30938

release-2.1: storage: Don't transfer leases to behind replicas in StoreRebalancer

135e0f6

danhhz


      changefeedccl: support a minimum frequency between resolved ts emits

The `WITH resolved` option now take an optional duration which is used
as a lower bound on the duration between emitted resolved timestamps. It
accepts any format that https://golang.org/pkg/time/#ParseDuration does.
(e.g. `WITH resolved='10s'`). If unspecified, the previous behavior of
emitting all resolved timestamps is used.

`resolved` is now the first `WITH` option to optionally take a value, so
this required a small refactor to make an enum from the bool specifing
whether the value was required.

Closes #30367

Release note (enterprise change): CHANGEFEEDs can now be configured with
a minimum duration between emitted resolved timestamps.

2382d83

knz


      sql: present the output of SHOW JOBS in sorted order

Requested by @rolandcrosby and @a-robinson.

Release note (sql change): The output of `SHOW JOBS` now reports
ongoing jobs first in start time order, then completed jobs in
finished time order.

92c0f7f

knz


      Merge pull request #31005 from knz/backport2.1-30917

release-2.1: sql: present the output of SHOW JOBS in sorted order

041f6bd

petermattis


      rpc/nodedialer: reset conn breaker after succesful connection

`Dialer.DialInternalClient` does not check the circuit breaker but
blindly attempts a connection and can succeed, leaving the system in a
state where there is a healthy connection to a node, but the circuit
breaker used for dialing is open. DistSQL checks for connection health
when scheduling processors, but the connection health check does not
examine the breaker. So DistSQL will proceed to schedule a processor on
a node but then be unable to use the connection to that node because
`Dialer.Dial` will return with a `breaker open` error. The code contains
a TODO to reconcile the handling of circuit breakers in the various
`Dialer` methods, but changing the handling is risky in the short
term. As a stop-gap, we reset the breaker after a connection is
successfully opened.

Fixes #29149

Release note: None

03cc365

petermattis


      sql: consider conn circuit breakers in distsql planning

Change `DistSQLPlanner.checkNodeHealth` so that it uses
`nodedialer.Dialer.ConnHealth` instead of `rpc.Context.ConnHealth`. The
former is the right method to be calling to check a node's connection
health.

Refactor `DistSQLPlanner.checkNodeHealth` into a `distSQLNodeHealth`
struct. This removed the need for `DistSQLPlannerTestingKnobs`.

Enhance `nodedialer.Dialer.ConnHealth` to mark connections as unhealthy
if the circuit breaker is open. This prevents DistSQL from planning
processors on such nodes.

Release note: None

dab62ae

petermattis


      Merge pull request #31014 from petermattis/backport2.1-30987

release-2.1: sql,rpc/nodedialer: improve distsql node health checks

00ee2a9

nvanbenschoten and knz


      pgwire: conn holds stmtBuf by value, not reference

Release note: None

f16ae14

nvanbenschoten and knz


      pgwire: conn holds writeBuffer by value, not reference

defc6cb

andreimatei and knz


      sql: minor renames

... and comments around the SessionRegistry.

Release note: None

57f10fd

andreimatei and knz


      sql, pgwire: cancel sessions better

Before this patch, `cancel session <foo>` would cause a connExecutor's
context to be canceled. This patch switches it to cancel the pgwire
connection's ctx instead (which was a parent of the connExecutor's ctx;
now the two are the same), as nature intended. For example, reading from
the network connection should stop ASAP after `cancel session`. It also
opens to door to guaranteeing that the client using the respective
session is unblocked in a timely manner (by the server closing the
network connection), although that is not currently implemented (pgwire
will still wait for the connExecutor to react to cancelation and finish
its task before closing the net.Conn).
This patch also eliminates what I presume to have been a crash when
someone tries to cancel a "session" created by the InternalExecutor. It
will now be a no-op.
This patch also moves a cancelation callback out of
connExecutor.ctxHolder. I'm trying to get rid of that guy, which is how
this patch came to be.

I've also added a TODO with some thoughts on a better design for session
cancelation, originally expressed in
#23861 (comment).
FWIW, I've tried to be the change, but gave up for the moment.

Release note: None

ff7eb11

knz


      sql: add error handling to the connExecutor constructor

Release note: None

09068e4

knz


      pgwire: add missing logging to the connection test code

Release note: None

59534ff

knz


      sql: properly ignore invalid chars in encoding names

Release note (bug fix): CockroachDB now properly ignores
non-alphanumeric characters in encoding names passed to
`convert_from()`, `client_encoding`, etc., for compatibility with
PostgreSQL.

3a0c833

knz


      sql,pgwire: fix a host of pgwire/sql interaction bugs

Prior to this patch, CockroachDB:

- incorrectly ignored the `extra_float_digits` client
  parameter.
- improperly reported `ISO` as default value for server
  status parameter and session var `DateStyle` (the pg
  default is `ISO, MDY`.
- failed to report `TimeZone`, `is_superuser` and
  `session_authorization` in the server status parameters.
- failed to report the correct (and client-provided) value of
  `application_name` in the server status parameters.
- did not report the correct defaults in `pg_catalog.pg_settings`.

This patch corrects these various bugs by:

- keeping the client-provided parameters as session defaults, to be
  picked up by RESET and pg_settings. (Prior to this patch, the
  client parameters were discarded after setting up the session
  initially.)
- extending the handling of session variables to support reporting
  their current value during connection intiialization. This way
  the server status parameter can report the session variables
  that were just initialized from the client-provided defaults.
- eliminating the ad-hoc constant list of "statusServerParams"
  maintained in pgwire/conn.go, which could dangerously go out of sync
  with the session variable defaults without anyone noticing.
- adding the missing reports for `is_superuser` and
  `session_authorization`.

Because CockroachDB now properly honors the value of
extra_float_digits provided by the client, any test that uses lib/pq
must now account for the fact that lib/pq automatically sets this
parameter to 2. The logic_test code is modified accordingly.

Because CockroachDB now properly reports the current TimeZone
parameter to the client during connection initialization, any test
that uses lib/pq must now account for the fact that lib/pq will load
that zone name into any parsed timestamp value coming from the
server. The tests that print out timestamps have been modified
accordingly.

As a side effect of this fix, it is now possible to set arbitrary
session variables in the client connection URL. This has another
benefit reported in the release notes below.

Release note (bug fix): CockroachDB now properly recognizes the value
of `extra_float_digits` provided by clients as a connection parameter.

Release note (bug fix): CockroachDB now properly recognizes two-part
values for the session variable and connection parameter `DateStyle`,
for compatibility with PostgreSQL.

Release note (bug fix): CockroachDB now reports all server status
parameters supported by PostgreSQL when setting up a session. This is
expected to improve compatibility with some drivers.

Release note (bug fix): CockroachDB now properly uses the
client-provided default values when using the RESET statement (or SET
... = DEFAULT).

Release note (bug fix): CockroachDB now properly fills the columns
`boot_val` and `reset_val` in `pg_catalog.pg_settings` for better
compatibility with PostgreSQL.

Release note (cli change): It is now possible to provide
initial/default values for any customizable session variable in the
client connection URL.

Release note (sql change): CockroachDB now supports more
customizations from PostgreSQL client drivers when initially setting
up the client connection.

7e35810

nvanbenschoten


      storage: only track forwarded proposals that were successfully appended

Fixes #30064.

Before this change, it was possible for a proposal that was forwarded to
a leader to be dropped instead of appended to its log. This was possible
if the leader had a `leadTransferee` set. This change avoids this issue
by verifying that the proposal was successfully added to the leaders log
before tracking it.

This will need to be backported to 2.1 and 2.0.

Release note: None

8f7e452

nvanbenschoten


      Merge pull request #31022 from nvanbenschoten/backport2.1-30990

release-2.1: storage: only track forwarded proposals that were successfully appended

161f232

jordanlewis


      sqlbase: don't permit PK columns in STORING list

Previously, users could create indexes that included PRIMARY KEY columns
in the STORING list. This was effectively a no-op and misleading to
users, since PRIMARY KEY columns are always stored in indexes regardless
of the STORING list. Separately, permitting this tickled a bug in the
descriptor accounting system.

Now, this behavior is prevented by returning an error to the user if
they try to do such a thing, with a helpful message explaining why it's
not allowed.

Release note (sql change): don't allow specifying STORING columns in
indexes that are already part of the index's corresponding PRIMARY
INDEX.

dfd65de

Commits on Oct 07, 2018

justinj and knz


      opt: support single-use CTEs

This commit introduces Cockroach-style single-use CTEs to the optimizer.
Such CTEs cannot be referenced more than once in a query.

Release note: None

00cb96c

justinj and knz


      opt: remove some unimplementedfs

These two checks are now exhaustive, meaning anything missing from them
going forward is an error.

Release note: None

abc449d

knz


      sql: bark at unused CTEs containing mutations

The SQL standard (and pg) specifies that CTEs containing mutations are
executed no matter how / how many times they are used in the remainder
of the query.

Meanwhile the heuristic planner + execution machinery is currently
unable to force the execution of CTEs containing mutations if they are
not otherwise used in the rest of the query.

However meanwhile they are also accepted silently -- i.e. CockroachDB
silently fails to do what the semantics require.

This patch changes the behavior to instead report an error "feature
not supported".

Release note (bug fix): CockroachDB now reports an unimplemented error
when a common table expression containing INSERT/UPDATE/UPSERT/DELETE
is not otherwise used in the remainder of the query.

95f2ec7

knz


      sql: do not silently ignore CTEs inside parentheses

Prior to this patch, CockroachDB silently ignored WITH clauses within
parentheses, so that the following SQL would return incorrect results:

```
CREATE TABLE t(x INT);
(WITH t AS (SELECT 1) TABLE t);
```

(note: `WITH ...` *without* parentheses did work fine)

This patch fixes this particular issue, and causes CockroachDB to
issue an "unimplemented error" in other cases not supported yet, for
example when there are multiple nested WITH clauses within
parentheses.

Release note (bug fix): CockroachDB does not silently ignore WITH
clauses within parentheses any more.

knz


      sql: properly report #10028 as an unimplemented feature

Release note: None

4587f92

knz


      sql: properly support CTEs inside views

Release note (sql change): CockroachDB now supports CTEs inside views.

d70baf4

knz


      Merge pull request #31021 from knz/backport2.1-30508-30813-30702

release-2.1: sql,pgwire: fix a host of pgwire/sql interaction bugs + perf fixes

bcef656

Commits on Oct 08, 2018

vivekmenezes


      sql: disable closed timestamp push in TestSchemaChangeAfterCreateInTxn

The schema change in TestSchemaChangeAfterCreateInTxn can occasionally
take more than 5 seconds and would get pushed by the read timestamp
cache by the closed timestamp mechanism. This effectively disables the
closed timestamp push for this test.

fixes #30492

Release note: None

8deedcf

benesch


      lint: reorganize package

The lint_test.go file was getting rather large. Split out the custom
"checkers" into their own files.

Release note: None

e0e1b93

benesch


      server,lint: fix misuse of hash.Hash.Sum

The hash.Hash interface is poorly designed and easily misused. Several
locations in our codebase were assuming that the Sum function returns
the hash of its input, when in fact it returns the concatenation of its
input with the hash of the empty string. See the comments within for the
full detail.s

This commit:

  1. Corrects the code in our web UI login to avoid leaking session
     secrets into the audit log. Since web UI login has only been
     released in betas, there are no backwards compatibility concerns
     here. Existing web sessions will simply be invalidated when the
     cluster is upgraded and users will need to re-login.

  2. Files a TODO about correcting the code in our password authentication. It
     is not a security risk there as the SHA-256 application was only
     intended to provided a fixed-length input to bcrypt. Fixing this
     code path is more difficult as it requires a migration for existing
     clusters, since this bug has been present since pre-1.0.

  3. Adds a linter to prevent future misuse.

Release note: None

0350a03

benesch


      storage: don't serve requests on RHS after a merge commits

Previously, it was possible for the leaseholder of the right-hand side
of a merge to erroneously serve requests after the merge committed. This
would occur when the RHS's watcher goroutine noticed that a merge
committed before the LHS on that store applied the merge commit trigger.
In this case, the watcher goroutine would unblock traffic, but that
traffic would be served normally because the RHS would not be marked as
destroyed. In effect, the watcher goroutine was assuming that the RHS
would always be marked as destroyed before it discovered that the merge
had committed, but there is no such guarantee.

The fix is to teach the watcher goroutine to mark the replica as
destroyed if the merge has committed before unblocking traffic.

With the fix in place, TestStoreRangeMergeRHSLeaseExpiration is no
longer flaky. For good measure, add a new test,
TestStoreRangeMergeWatcher, which more reliably reproduces the
consistency violation.

Fix #30349.

Release note: None

78953b4

couchand


      ui: visually embed node lists in container

Release note: None

df8658b

couchand


      ui: add bar chart for node list memory & capacity

Fixes: #29373
Release note (admin ui change): Adds a bar chart to the memory and
capacity usage columns on the node list.

d2a7be9

couchand


      ui: sort memory & capacity columns by percent

Release note (admin ui change): Sorting by the memory or capacity
usage column on the node list now sorts by percentage used.

d5a6932

couchand


      ui: fix node list bar height in Chrome

Chrome and Firefox have different rules about CSS selectivity
for SVG elements, apparently.  This is a hack, we should find
a simpler way to handle this (maybe changing the conflicting
classes).

Release note: None

8b82c05

vivekmenezes


      Merge pull request #31056 from vivekmenezes/backport2.1-30997

backport-2.1: sql: disable closed timestamp push in TestSchemaChangeAfterCreateInTxn

7ff3d14

knz and jordanlewis


      sql: properly rename stored columns

Prior to this patch ALTER TABLE RENAME COLUMN did not properly rename
stored column in indexes. This patch fixes it.

Release note (bug fix): CockroachDB now properly supports renamining a
column that's also stored in an index.

1c5ac75

jordanlewis


      Merge pull request #31032 from jordanlewis/backport2.1-31019

backport-2.1: sqlbase: don't permit PK columns in STORING list

e73a623

jordanlewis


      Merge pull request #31074 from jordanlewis/backport2.1-31010

release-2.1: sql: properly rename stored columns

1dd2fd0

knz


      Merge pull request #30218 from knz/backport2.1-30210

release-2.1: cli: hint the syntax of client commands in `start` output

8b5891f

a-robinson


      server: Fix EnqueueRange to respect its NodeID parameter

This got broken by some sloppy refactoring to make the method use
iterateNodes rather than a more one-off approach to forwarding the
request to nodes using the isLiveMap.

Release note (bug fix): Fix the _admin/v1/enqueue_range debug endpoint
to always respect its node_id parameter.

5a916f5

benesch


      Merge pull request #31058 from benesch/backport2.1-30986

release-2.1: storage: don't serve requests on RHS after a merge commits

petermattis


      gossip: use "n%d" instead of "node %d" in log messages

This normalizes node IDs in gossip log messages with the rest of the
code base.

Release note: None

7c3289e

petermattis


      gossip: avoid sending unnecessary client to server infos

Delay registration of the callback for triggering client to server info
propagation until we've received the first response from the server
which will contain the server's high water stamps.

Add a test (currently skipped) which demonstrates the problem the
most-distant calculation.

See #30895

Release note: None

acb3883

petermattis


      gossip: propagate most-distant marker infos on every connection

Whenever a gossip connection is created, propagate the most-distant
marker infos (i.e. `node:<id>` infos) to the remote node. These infos
are used for the `mostDistant` calculation and need to accurately
reflect the number of hops to a node.

Note that this is a bandaid. Using the info hops value to calculate the
distance to a node does not work correctly unless infos are always
propagated. The high water stamps mechanism that avoids sending an info
that the remote already has can allow the info hops value to remain
higher or lower than the actual value. Consider the following gossip
network:

  1 -> 2
  1 -> 3
  2 -> 3
  3 -> 4

When n1 gossips its `node:1` info it has multiple paths to get to n4. We
can either see `1 -> 2 -> 3 -> 4` or `1 -> 3 -> 4`. Depending on the
path the hops value will differ and the path is timing dependent. Even
worse, consider what happens when the network changes. For example,
start with the network `1 -> 2 -> 3 -> 4`. The network then changes to:

  1 -> 3
  2 -> 3
  3 -> 4

The `node:<id>` keys have a TTL of 2h. In the initial configuration, n2
is considered to be 1 hop distant from n1. But after the
reconfiguration, n2 is in reality 2 hops distant from n1, yet the
`node:2` key living on n1 will not have been updated because it hasn't
changed and the new hops value is higher. The new `gossip-clients:*`
keys give us a way out of this complexity as each node can compute the
shortest path to every other node, but those keys are not generated by
v2.0.x of CockroachDB so we'll have to wait to use them for that
purpose.

Fixes #30895

Release note: None

e458381

ridwanmsharif


      sql: remove protobuf output from `show zone configurations`

This shouldn't be in the output for the 2.1 release,
the tests that use this to verify zone configs now
can use the `crdb_internal.zones` table.

Release note (sql change): output of `SHOW ZONE CONFIGURATIONS`
and `SHOW ZONE CONFIGURATION FOR` now only shows the zone name
and the SQL representation of the config.

19edd83

solongordon


      distsql: eagerly evaluate constant expressions

exprHelper now evaluates constant subexpressions upon initialization,
rather than re-evaluating them every time `eval` or `evalFilter` is
called, and thus avoids a hefty per-row cost for expensive expressions.
For example, the SELECT query in the example below now returns in ~30ms
rather than ~10s.

```
CREATE TABLE t (x int[]);
INSERT INTO t SELECT array[x] FROM generate_series(1, 10000) AS x;
SELECT * FROM t WHERE x = (SELECT array_agg(generate_series(1,10000)));
```

Fixes #30167

Release note: None

abe0966

couchand


      Merge pull request #31070 from couchand/backport2.1-30982

release-2.1: ui: add bar chart for node list memory & capacity

73476c1

a-robinson


      ui: Add debug page for the EnqueueRange admin endpoint

It's probably the worst-constructed page this UI has ever seen, but I
managed to get something working. Feedback appreciated, but keep in mind
that we'd like to squeeze something in for 2.1 (unless there are
security-related reasons to keep it out, in which case the endpoint
itself would be a concern moreseo than this page), since without this
sort of page using the endpoint is pretty tough.

The only issue I still have is that 400 errors don't display any detail
about the error, just "Bad Request", and I'm not sure how to fix that.
It's far from a show stopper, though.

Release note (admin ui change): Added a debug page with a form that lets
you manually enqueue a range in one of the various store-level replica
queues on the store of your choice. For advanced users only.

92b701a

vilterp and a-robinson


      ui: enqueue range: fix linter errors

Release note: None

636af88

vilterp and a-robinson


      ui: enqueue range: fix up table output

Release note: None

984c710

vilterp and a-robinson


      ui: enqueue range: put queue name options in dropdown

Release note: None

a-robinson


      ui: enqueue range: fix style nits

Release note: None

ede6d64

petermattis


      Merge pull request #31088 from petermattis/backport2.1-31025-31042

release-2.1: gossip: propagate most-distant markers on every connection

ec47877

ridwanmsharif


      Merge pull request #31089 from ridwanmsharif/backport2.1-30985

release-2.1: sql: remove protobuf output from `show zone configurations`

5e12472

mjibson


      Merge pull request #31057 from benesch/backport2.1-30859

release-2.1: server,lint: fix misuse of hash.Hash.Sum

64990bb

mjibson


      Merge pull request #30935 from mjibson/backport2.1-30932

release-2.1: security: use same error message for user DNE and bad password

21bf597

a-robinson


      Merge pull request #31092 from a-robinson/backport2.1-31035

release-2.1: ui: Add debug page for the EnqueueRange admin endpoint

dc0e73c

jordanlewis


      sql: fix planning of wrapped local nodes

Previously, wrapped local nodes that contained omitted columns could
crash the database in various ways. DistSQL physical planning wasn't
taking into account the omitted state of columns for wrapped nodes.

This commit fixes the issue.

Release note: None

a6465ba

solongordon


      Merge pull request #31091 from solongordon/backport2.1-30216

release-2.1: distsql: eagerly evaluate constant expressions

ad11a25

a-robinson


      Merge pull request #31087 from a-robinson/backport2.1-31036

release-2.1: server: Fix EnqueueRange to respect its NodeID parameter

5cd3520

danhhz


      Merge pull request #31008 from danhhz/backport2.1-30974

release-2.1: changefeedccl: support a minimum frequency between resolved ts emits

324dfeb

vilterp


      sql: include distsql queries in `sql.{service,exec}.latency` metrics

Previously, the `sql.{service,exec}.latency` histogram metrics only
included local queries; distributed queries' latencies were recorded in
the `sql.distsql.{service,exec}.latency` metrics.

However, the UI had assumed that the `sql.service.latency` included all
SQL statements both local and distributed, and therefore had a graph of
it on the Overview dashboard, as well as having its most recent value
in the sidebar of the time series area.

With this change, the latencies of all statements are recorded in
`sql.{service,exec}.latency`, changing those metrics to mean what we had
assumed they meant. DistSQL queries will no longer be left out of this
prominent graph and sidebar metric.

Release note (admin ui change): The "service latency: {90,99}th
percentile" graphs on the Overview and SQL dashboard, as well as the P50
and P99 latency numbers in the time series area sidebar, now reflect
latencies of both local and distributed queries. Previously, they only
included local queries.

eaecee6

vilterp


      ui: remove DistSQL latency graphs from SQL dashboard

These are not very useful now that the execution engines have been
merged.

Release note: none

7d4b190

Commits on Oct 09, 2018

jordanlewis


      Merge pull request #31097 from jordanlewis/backport2.1-30942

release-2.1: sql: fix planning of wrapped local nodes

199229e

couchand


      ui: open docs links in new tab

Closes: #24780
Release note (admin ui change): Links to documentation pages
now open in a new tab.

eeff56d

couchand


      sql: only quantize counts when scrubbing stmt stats

Fixes #30369.
Release note: None

337efb0

couchand


      Merge pull request #31136 from couchand/backport2.1-31099

release-2.1: sql: only quantize counts when scrubbing stmt stats

8e1962c

changangela


      sql: rename crdb_internal.ranges column names

Release note: None

70e2654

danhhz


      changefeedccl: support TIMESTAMP and DECIMAL with the avro format

Timestamp is relatively straightforward and uses the corresponding
logical type.

The avro spec also has a logical type for decimals, but ours cover a
number of cases that it doesn't (NaN, Inf, per-datum precision and
scale). I think the right answer for them is to make this a union with a
`crdb.decimal` logical type that has a string representation.
Unfortunately, we can't do that without some surgery to the avro library
we use and it seems too late in the game to do that. For now, leave them
unsupported.

A number of column types are still unsupported with avro, but this is
probably a good enough core that we can mention the experimental
support.

Release note (enterprise change): CHANGEFEEDs now have limited and
experimental support for the avro format.

0ccf1be

changangela


      Merge pull request #31138 from changangela/sql-change-ranges-columnna…

…mes-release-2.1

release-2.1: rename crdb_internal.ranges column names

59721d8

danhhz


      changefeedccl: fix TestAvroSchema/DECIMAL flake

The precision is really meant to be in [1,10], but it sure looks like
there's an off by one error in the avro library that makes this test
flake if it picks precision of 1.

Release note: None

1e8cb49

vilterp


      Merge pull request #31116 from vilterp/backport2.1-31098

release-2.1: sql, ui: include distsql queries in `sql.{service,exec}.latency` metrics

823c43d

danhhz


      Merge pull request #31143 from danhhz/backport2.1-31104

release-2.1: changefeedccl: support TIMESTAMP and DECIMAL with the avro format

da40442

mrtracy


      cdc: Test for feed falling behind table GC

Constructs a test which verifies that Changefeeds exit with an error if
they fall behind the GC TTL of the table data being read.

In order to test this correctly, a new "ManuallyEnqueueSpan()" function
has been added to the Store object, intended for internal tests with
access to the Store to manually force any replicas in the provided span
through a specific replica queue.

Release note: None

4ac9703

mrtracy


      cdc: Test for falling behind schema TTL

Add a test that ensures that changefeeds properly exit if they fall far
enough behind that schema information has been lost due to the GC TTL
(that is, a historical row version can no longer be read because the
schema at its timestamp has been garbage collected).

I have also discovered why the sister test (for the table TTL, not the
schema) required a 3 second sleep: the GC queue enforces that replicas
must have an appropriately high "score" before being GCed, even when the
"shouldQueue" process is skipped. I have swapped to using the
"ManuallyEnqueue" method to force the GC, and instead use the already
existing GCRequest request type to force a GC of the table; as a result
the tests no longer require any sleep.

Resolves #28644

Release note: None

93efdce

mrtracy


      cdc: Fix Sink Error Retries + Chaos Test

Fixes a few related bugs that were causing the CDC kafka chaos test to
fail-

The `retryableSinkError` returned from kafkaSink was not being returned
as a pointer, meaning it was not being detected correctly. Additionally,
retryable connection errors could be thrown while attempting to connect
the sink initially, but these were not being marked as retryable.

If the kafka error occurs on a remote node in the DistSQL operation, it
is marshalled over the wire as a `pgerror.Error`, meaning it can no
longer be detected with type assertions. We now check for this in
`isRetryableError`.

Finally, the CDC chaos test itself was not properly silencing possible
errors from `confluent destroy`.

Fixes #29243

Release note: None

24c4d2a

couchand


      ui: add license header to all oss code

Fixes: #23376
Release note: None

5e8bf87

mrtracy


      Merge pull request #31161 from mrtracy/backport2.1-30335-31020-31037

backport-2.1: cdc: GC TTL tests + Fix Sink Error Retries

2d468e5

mrtracy


      cdc: Support Schema Changes with Backfill

Changefeeds can now correctly continue when watched tables are altered
in ways that require a backfill operation.

To represent a backfill through the changefeed, a full scan of
the watched tables is performed at the timestamp when the table
descriptor for the finalized schema change is detected (when the
modified columns become public).

Note that, due to the backfill process writing rows in the background,
the changefeed will see numerous "no-op" writes, where each backfilled
row is written with no apparent changes. However, this does not
violate any correctness guarantees, it is just inefficient.

Additionally, note that this commit changes the behavior of initial
scans; previously, an initial scan would return the latest value for
each key at the timestamp of the key. However, in order to capture
schema changes that occurred in between the CHANGEFEED creation and the
last write of each key, we now treat the initial scan as a backfill and
output each key at the statement time. We have temporarily modified one
test to allow this to be committed, re-enabling it is tracked by #31110.

Release note: Changefeeds now continue running when watched tables
are ALTERed in ways that require a backfill.

Commits on Oct 10, 2018

mrtracy


      Merge pull request #31165 from mrtracy/backport2.1-30902

backport-2.1: cdc: Support Schema Changes with Backfill

163ebe3

nvanbenschoten


      storage: reduce commit index on quiescence heartbeat, don't drop them

Informs #30064.

53dc851 introduced logic to drop quiescence heartbeats to Replicas that
are not caught up to the leader. This turns out to be an issue in cases
where:
- a Range is quiescent
- a Replica needs a heartbeat to join a Raft group
- that Replica's node is unable to update its liveness record because
doing so depends on the Replica joining the Raft group
- so the Replica never gets any heartbeats and never joins the group

This change adjusts the logic. Instead of dropping these heartbeats, the
leader Replica now sends them, but with a reduced commit index and with
Quiesce equal to false.

Release note: None

8f55a8e

nvanbenschoten


      storage: remove unreachable code in quiesceAndNotifyLocked

Release note: None

1383d09

jordanlewis


      sql: always include filters in EXPLAIN

Previously, EXPLAIN only included filters and join conditions when given
the VERBOSE option. This proved to be confusing for debugging, as the
heuristic of checking a scan for spans doesn't necessarily mean that
scan has enough indexes to be efficient. Unconditionally showing the
filter condition will make this a lot more obvious.

Release note (sql change): EXPLAIN now always shows filter and join
conditions.

733619d

jordanlewis


      Merge pull request #31186 from jordanlewis/backport2.1-31168

release-2.1: sql: always include filters in EXPLAIN

71e165d

couchand


      Merge pull request #31162 from couchand/backport2.1-31157

release-2.1: ui: add license header to all oss code

30d91b6

solongordon


      distsql: don't use sortTopK when filter is present

The sorter was producing incorrect results when both a limit and a
filter were applied. We can't use sortTopK in this case since some
results may be filtered out in post-processing. Note this scenario is
somewhat rare because typically the filter would be pushed down below
the sort. The issue was observed when selecting from the result of a
SHOW TRACE.

Also removed a bit of dead code.

Fixes #31163

Release note: None

cf5020d

vivekmenezes


      sql: fix index backfill read before GC TTL threshold error

Before this change the historical read timestamp would remain fixed
for the duration of the index backfill. But if the index backfill
takes a long time or keeps restarting the timestamp can exceed the
GC window (configurable and defaults to 24hrs). Sometimes customers
are told to set the GC window to as low as 10m to reclaim disk space.

It was never clear what was the value of keeping this indefinitely
fixed. In fact it's quite likely it's worse to keep it fixed,
because any UPDATE to a row is likely to trigger redoing the index
backfill for the subspan containing the row at the current time.

This change marks the GC TTL threshold error as a non-permanent error.
On hitting the error, the backfill will be restarted at the last
checkpoint after picking a new historical read timestamp.

fixes #29192

Release note (sql change): fix schema change rollback caused by
GC TTL threshold error.

vilterp


      server: fix crash of data distribution endpoint after dropping a table

The crash was caused by trying to get the zone config with `SHOW ZONE
CONFIGURATION FOR TABLE` for a table that had been dropped, producing an
unexpected error.

Release note: None

4e3e6dd

vilterp


      server: fix benchmark of data distribution endpoint

Release note: None

a750d1e

vilterp


      ui: data distribution: grey out dropped tables

Also add a blurb under the zone config list explaining what greying out
means.

Release note: None

d793002

vilterp


      ui: data distribution: tweak zone config styling

Just show the SQL, not the cli specifier, since we don't use CLI
specifiers anymore.

Release note: None

9e73248

vilterp


      ui: refactor: provide default className for Loading component

Most usages pass `loading-image loading-image__spinner-left` to the
className prop; make that the default and remove the prop for those
usages.

Release note: None

2acf94c

vilterp


      ui: refactor: make spinner default image to Loading component

...and remove all that prop from all the usages that passed it.

Release note: None

f9adb60

couchand


      Merge pull request #31132 from couchand/backport2.1-31105

release-2.1: ui: open docs links in new tab

7038fe7

vilterp


      Merge pull request #31207 from vilterp/backport2.1-31117

release-2.1: server, ui: fix data distribution endpoint

a357884

couchand and vilterp


      ui: extract infobox from login page

Release note: none

1880d47

couchand and vilterp


      ui: add infobox to debug index page

Release note: None

b3049dd

couchand and vilterp


      ui: move top-level debug pages to panel above

Release note: None

couchand and vilterp


      ui: move license type to top of debug page

Release note: None

6297d7e

vilterp


      ui: add link to debug pages in sidebar

Release note: None

a9deab3

vilterp


      ui: make entire logout icon clickable

Release note: None

bb15ba7

couchand and vilterp


      ui: clean up debug pages link

Release note: None

9c907d9

couchand and vilterp


      ui: update gear icon

Release note: None

b1ba914

couchand and vilterp


      ui: extract classnames call in iconlink

Release note: None

324b513

couchand and vilterp


      ui: narrow spacing on debug pages link

Release note: None

a2af844

vilterp


      Merge pull request #31225 from vilterp/backport2.1-31218

release-2.1: ui: surface debug pages

aec17b1

couchand and vilterp


      ui: add empty state to databases page

Fixes: #27620
Fixes: #28234
Release note (admin ui change): Improve the view of databases with no tables.

9b9f3b1

Commits on Oct 11, 2018

vivekmenezes


      Merge pull request #31153 from vivekmenezes/backport2.1-30648

backport-2.1: sql: fix index backfill read before GC TTL threshold error

48fb937

couchand


      Merge pull request #31231 from vilterp/backport2.1-31137

release-2.1: ui: add empty state to databases page

b445f61

vilterp


      ui: avoid using font which isn't loaded yet in alert banner

The "we're having trouble connecting to the cluster" banner, while
already annoying, also often looks janky because it's in Times New
Roman. This happens because it's styled to use Lato-Bold, which isn't
used much in the UI and as such usually hasn't been loaded at the point
that the banner appears. It can't be loaded when the banner appears
because the UI can't connect to the cluster.

This change uses Lato-Regular with `font-face: bold` applied, which
looks almost the same but is always loaded so we don't see the banner in
Times New Roman.

Release note: None

80b84d1

vilterp


      ui: allow SortedTable and SortableTable to expand and collapse rows

if optional "expandable config" is provided.

Release note: None

54b576a

vilterp


      ui: jobs page: use expandable table

Release note (admin ui change): Update jobs table to make each row
expandable, allowing the user to see the error message for failed jobs.

b5b0459

vilterp


      ui: jobs page: remove background colors based on job state

With the icons, these add more visual distraction than they're worth.

Release note: None

4cdfc7b

vilterp


      ui: jobs page: add icons to indicate job state

Green checkmark for completed, red exclamation point for failed.

Release note: None

b885f83

vilterp


      ui: lint: fix indentation in jobs page

To use

```
return (
  <div>
    ...
  </div>
);
```

Instead of

```
return <div>
 ...
</div>;
```

Release note: None

c1361a2

vilterp


      ui: lint: fix inconsistent indentation in sqlhighlight.styl

Release note: None

768ae92

vilterp


      ui: jobs page: update color of progress bar

To match palette

Release note: None

18863a8

vilterp


      ui: refactor: pages/jobs.styl => jobs/index.styl

Release note: None

vilterp


      Merge pull request #31236 from vilterp/backport2.1-31228

release-2.1: ui: avoid using font which isn't loaded yet in alert banner

c0e5a56

vilterp


      Merge pull request #31237 from vilterp/backport2.1-29450

release-2.1: ui: make rows in jobs table expandable to see details

d0aef0c

benesch and tschottdorf


      storage: deflake TestStoreRangeMergeWatcher

This test could deadlock if the LHS replica on store2 was shut down
before it processed the split at "b". Teach the test to wait for the LHS
replica on store2 to process the split before blocking Raft traffic to
it.

Fixes #31096.
Fixes #31149.
Fixes #31160.
Fixes #31167.

Release note: None

5dad79a

nvanbenschoten and tschottdorf


      storage: deflake TestStoreSplitRangeLookupRace

Fixes #30784.

Release note: None

580dc23

tschottdorf


      github-pull-request-make: fail on hanging tests

Inspired by:
#30986 (review).

Release note: None

4a86758

tschottdorf


      Merge pull request #31248 from tschottdorf/backport2.1-31215

backport-2.1: storage: deflake TestStoreRangeMergeWatcher

8f112dc

vilterp


      ui: scope table padding under debug index panel section class

...so it doesn't apply to the jobs page and other pages.

Release note: None

cd18046

nvanbenschoten


      Merge pull request #31174 from nvanbenschoten/backport2.1-31112

release-2.1: storage: reduce commit index on quiescence heartbeat, don't drop them

67c66e2

vilterp


      Merge pull request #31262 from vilterp/backport2.1-31244

release-2.1: ui: scope table padding under debug index panel section class

974ce6a

tschottdorf


      Merge pull request #31253 from tschottdorf/backport2.1-31063

backport-2.1: github-pull-request-make: fail on hanging tests

8c6f525

tschottdorf


      storage: avoid spurious nodeDialer logs

breaker.Success() retrieves its count from an underlying windowed
counter that resets every 10 seconds. As a result, spurious "established
connection to nX" messages would pop up every 10s.

It turns out it's annoyingly difficult to properly gauge whether an
established connection is the first of its kind. In trying to do so, I
became unconvinced that this is even worth it, so I ended up removing
it. I similarly simplified the logging in the unsuccessful case (though
ConsecutiveFailures is actually not a windowed counter and presumably
worked correctly).

I also downgraded the message in the error case to Infof, as it's
expected to see it when nodes are restarted (which is a routine
operation and nothing to worry about). We want to keep the Warning level
for true warnings.

Release note: None

a62ec4e

tschottdorf


      nodedialer: Reset() breaker only when it is tripped

This is slightly better since it only resets the breaker state (and with
it the underlying counters) when it's tripped.

This shouldn't result in any external changes but made the problem fixed
in #31017 much more visible than it would've otherwise been.

Release note: None

999f963

tschottdorf


      Merge pull request #31249 from tschottdorf/backport2.1-30816

backport-2.1: storage: deflake TestStoreSplitRangeLookupRace

1d70a80

justinj


      opt: normalize ANY = ... to IN ...

This commit adds two rules which, in concert, allow a filter of the form

    x = ANY ARRAY[...]

to constrain a scan on an index of x.

It's unclear to me if this is a win in the general case, but addressing
the conversation from a couple days ago where we discussed enabling a
placeholder ARRAY to be used instead of a tuple to possible make better
use of the plan cache.

Release note: None

b3d13e3

knz


      sql: capture internal/assertion errors

Prior to this patch, assertion/internal errors were "merely" reported
to the client with pg error code CodeInternalError and nothing would
happen further. This was causing 2 problems:

- it was unclear for the user what to do with this information.
- the occurrences were not further collected and could not be used to
track bugs automatically.

This patch improves the situation as follows:

- a new constructor `pgerror.NewAssertionErrorf` is introduced, which:
- gives a hint to the user about submitting the error as a bug with
details.
- populates the InternalCommand with a shortened stack trace.

- the new constructor is used instead of
`pgerror.NewErrorf(pgerror.CodeInternalError` throughout the SQL
layer.

- the reporting mechanism now detects CodeInternalError errors,
collects then reports them with other node statistics.

Before:

```
root@127.0.0.1:56399/defaultdb> select crdb_internal.force_assertion_error('woo');
pq: crdb_internal.force_assertion_error(): programming error: woo
```

After:

```
root@127.0.0.1:63733/defaultdb> select crdb_internal.force_Assertion_error('woo');
pq: crdb_internal.force_assertion_error(): programming error: woo
DETAIL: builtins.go:2805,eval.go:3547,expr.go:194,processors.go:378,processors.go:745,values.go:125,base.go:170,processors.go:768,flow.go:596,distsql_running.go:261,distsql_running.go:768,conn_executor_exec.go:983,conn_executor_exec.go:825,conn_executor_exec.go:403,conn_executor_exec.go:95,conn_executor.go:1189,conn_executor.go:466,conn.go:309
HINT: You have encountered an error inside CockroachDB. Don't worry! Your data is likely safe.

Please report this error with details at:
https://github.com/cockroachdb/cockroach/issues/new/choose
or
support@cockroachlabs.com

Thank you!
```

Release note (sql change): CockroachDB will now hint that internal
errors should be reported as bugs by users. Additionally, internal
errors are now collected internally and submitted (anonymized) with
other node statistics when statistic collection is enabled.

9a60220

a-robinson


      sqlmigrations: Downgrade log message from error to info

Fixes #31269

Release note: None

3d9f4ba

asubiotto


      sql: make DISTSQL option to EXPLAIN ANALYZE optional

Reduce user surprise by not requiring a DISTSQL option in EXPLAIN
ANALYZE.

Release note (sql change): EXPLAIN ANALYZE <statement> is now a valid
equivalent of EXPLAIN ANALYZE (DISTSQL) <statement>

bee759e

knz


      Merge pull request #31051 from knz/backport2.1-30858-30939-31003-3100…

…4-31007

release-2.1: assorted CTE fixes

0baee4f

knz


      Merge pull request #31272 from knz/backport2.1-31261

release-2.1: sql: capture and report internal/assertion errors

92ef8d7

knz


      sql: fix the schema type check for INSERT/UPSERT

The types of the input data for a mutation (the data source clause for
INSERT/UPSERT, the result of SET RHS for UPDATE) must match the types
of the columns being mutated.

Prior to this patch, this check was properly done early by UPDATE,
however it was not done at the right time for INSERT/UPSERT.

The code for INSERT/UPSERT was relying on an assertion check very deep
down in the encoding of datums to KV values. This assertion check was
never meant to be an input validation check, but because Go does not
really give us semantic assertions nobody noticed the check was
abused.

This patch clarifies the situation by introducing
`pgerror.NewAssertionErrorf` and using it there as appropriate. Later
patches can introduce more uses of pgerror.NewAssertionErrorf where
appropriate (e.g. everywhere we already use CodeInternalError).

In addition, the proper check for INSERT/UPSERT is introduced at the
right place, early during semantic analysis.

Release note (bug fix): CockroachDB will no longer fail in unexpected
ways or write invalid data when the type of input values provided to
INSERT/UPSERT does not match the type of the target columns.

c35092b

knz


      sql: produce the right pg error codes for column constraint failures

Release note: None

f750e45

knz


      sql: ensure UPDATE checks column constraints before CHECK constraints

This is required for postgres compatibility: the CHECK expressions
must be able to assume that the column data is already valid according
to the column type and nullability constraint.

The code for INSERT/UPSERT was already valid, only UPDATE needed to be
fixed. However this patch provides the corresponding test for all 3
mutations.

Release note (bug fix): UPDATE now verifies the column constraints
before CHECK constraints, for compatibility with PostgreSQL.

cd20f42

knz


      sql: add missing mutation check tests

The mutations must validate column constraints before they perform FK
checks. There were no clear tests for this. This patch adds the
missing tests.

Release note: None

99a7bbb

knz


      sql: prevent UPDATE RETURNING from seeing non-public columns

Prior to this patch, it was possible to use a newly added but not yet
visible column in UPDATE RETURNING. This patch fixes it.

INSERT/UPSERT/DELETE were not affected, however this patch adds the
corresponding tests for all 4 mutation statements -- this aspect
was not tested before.

Release note (bug fix): It is not possible any more to use
not-fully-added-yet columns in the RETURNING clause of UPDATE
statements.

85be94b

knz


      sql: fix the generation of an error message

Release note: None

cd9e57f

knz


      sql: add a missing comment

Release note: None

da762c0

knz


      sql: prevent INSERT DO UPDATE from silently setting computed columns

Release note (bug fix): CockroachDB does no longer incorrectly and
silently accept a computed column on the left side of the assignment
in a ON CONFLICT DO UPDATE SET clause.

04e62d1

knz


      sql: improve an error message

Release note: None

knz


      sql: prevent INSERT DO UPDATE SET from assigning mutation columns

This is a band aid, a refactoring of the upsert pipeline is required
for a more robust solution.

Release note (bug fix): CockroachDB does no longer incorrectly and
silently accept a not-fully-added-yet column on the left side of the
assignment in a ON CONFLICT DO UPDATE SET clause.

f6c884d

tschottdorf


      Merge pull request #31252 from tschottdorf/backport2.1-31017

backport-2.1: storage: avoid spurious nodeDialer logs

062109d

justinj


      Merge pull request #31271 from justinj/backport2.1-30973

backport-2.1: opt: normalize ANY = ... to IN ...

6c22a8d

tschottdorf


      storage: de-flake TestStoreRangeWaitForApplication

Errant requests could sneak in and confuse the lease counter. Lock down
the proposals for r2 while we're relying on a predictable
LeaseAppliedIndex.

Fixes #30866.

Release note: None

6d94b10

solongordon


      Merge pull request #31190 from solongordon/backport2.1-31189

release-2.1: distsql: don't use sortTopK when filter is present

changangela


      sql: add merge joiner benchmark and memory account shrinking optimiza…

…tion

Merge joiner benchmark comparison against `release-2.1`

```
name                           old time/op    new time/op    delta
MergeJoiner/InputSize=0-8        4.31µs ±10%    4.33µs ±15%      ~     (p=0.968 n=10+9)
MergeJoiner/InputSize=4-8        8.34µs ± 3%    8.10µs ± 4%      ~     (p=0.074 n=8+9)
MergeJoiner/InputSize=16-8       16.3µs ± 2%    11.1µs ± 5%   -31.87%  (p=0.000 n=9+9)
MergeJoiner/InputSize=256-8       190µs ± 2%      85µs ± 2%   -55.04%  (p=0.000 n=10+8)
MergeJoiner/InputSize=4096-8     2.96ms ± 2%    1.28ms ± 2%   -56.86%  (p=0.000 n=10+10)
MergeJoiner/InputSize=65536-8    49.0ms ± 5%    20.6ms ± 1%   -57.88%  (p=0.000 n=10+10)

name                           old alloc/op   new alloc/op   delta
MergeJoiner/InputSize=0-8        6.42kB ± 0%    6.65kB ± 0%    +3.49%  (p=0.000 n=10+10)
MergeJoiner/InputSize=4-8        9.50kB ± 0%    9.72kB ± 0%    +2.36%  (p=0.000 n=10+10)
MergeJoiner/InputSize=16-8       9.50kB ± 0%    9.72kB ± 0%    +2.36%  (p=0.000 n=10+10)
MergeJoiner/InputSize=256-8      32.5kB ± 0%    32.8kB ± 0%    +0.69%  (p=0.000 n=10+10)
MergeJoiner/InputSize=4096-8      401kB ± 0%     401kB ± 0%    +0.06%  (p=0.000 n=9+10)
MergeJoiner/InputSize=65536-8    6.30MB ± 0%    6.30MB ± 0%    +0.00%  (p=0.000 n=9+10)

name                           old allocs/op  new allocs/op  delta
MergeJoiner/InputSize=0-8          14.0 ± 0%      15.0 ± 0%    +7.14%  (p=0.000 n=10+10)
MergeJoiner/InputSize=4-8          16.0 ± 0%      17.0 ± 0%    +6.25%  (p=0.000 n=10+10)
MergeJoiner/InputSize=16-8         16.0 ± 0%      17.0 ± 0%    +6.25%  (p=0.000 n=10+10)
MergeJoiner/InputSize=256-8        46.0 ± 0%      47.0 ± 0%    +2.17%  (p=0.000 n=10+10)
MergeJoiner/InputSize=4096-8        526 ± 0%       527 ± 0%    +0.19%  (p=0.000 n=10+10)
MergeJoiner/InputSize=65536-8     8.21k ± 0%     8.21k ± 0%    +0.01%  (p=0.000 n=10+10)

name                           old speed      new speed      delta
MergeJoiner/InputSize=4-8      7.67MB/s ± 3%  7.91MB/s ± 4%      ~     (p=0.070 n=8+9)
MergeJoiner/InputSize=16-8     15.7MB/s ± 2%  23.0MB/s ± 5%   +46.89%  (p=0.000 n=9+9)
MergeJoiner/InputSize=256-8    21.6MB/s ± 2%  48.0MB/s ± 2%  +122.41%  (p=0.000 n=10+8)
MergeJoiner/InputSize=4096-8   22.1MB/s ± 2%  51.3MB/s ± 2%  +131.81%  (p=0.000 n=10+10)
MergeJoiner/InputSize=65536-8  21.4MB/s ± 5%  50.8MB/s ± 1%  +137.16%  (p=0.000 n=10+10)
```

Release note: None

7822e30

jordanlewis and tschottdorf


      sql: attempt to deflake distsql physical planner tests

Make sure the range cache is populated before verifying things about it.

Release note: None

be62f2d

a-robinson


      Merge pull request #31276 from a-robinson/backport2.1-31270

release-2.1: sqlmigrations: Downgrade log message from error to info

6cd3976

jordanlewis


      Merge pull request #31285 from tschottdorf/backport2.1-31239

backport-2.1: sql: attempt to deflake distsql physical planner tests

df65f23

Commits on Oct 12, 2018

knz


      sql: enable client to override index selection for DELETE and UPDATE

Prior to this patch, if the automatic index selection for DELETE or
UPDATE was inadequate, there was no way to override it like it is
possible for SELECT/INSERT/UPSERT.

This patch fixes this by extending and supporting the syntax for
DELETE and UPDATE:

```
DELETE FROM tbl@idx ...
UPDATE tbl@idx SET ...
```

Release note (sql change): it is now possible to force a specific
index for DELETE or UPDATE.

lhirata and knz


      Update the `unique_rowid` description

Release notes: None

4d7987b

lhirata and knz


      Regenerate functions docs

Release notes: None

2396a57

mberhault and knz


      cli: minimize --certs-dir flag description.

Certificate setup has expanded to the point where details cannot be
explained sufficiently here. Users should refer to documentation instead.

Release note: none

64452a0

knz


      Merge pull request #31279 from knz/backport2.1-31012

release-2.1: sql: enable client to override index selection for DELETE and UPDATE

73e7ce2

jordanlewis


      sql: better memory accounting in windower

Previously, the size of the bucket key wasn't taken into account. Also,
now nil out the partition map when closing the windower to ensure memory
is freed in a timely manner.

Release note: None

896a39d

jordanlewis


      sql: fix flaky TestShowQueries

There was a potential data race, and an incorrect check for a common
query (it's SHOW CLUSTER SETTING version, not capital VERSION).

Release note: None

a1165a6

jordanlewis


      Merge pull request #31302 from jordanlewis/backport2.1-30776-31241

backport-2.1: sql: better memory accounting in windower; fix flaky TestShowQueries

6c689c7

bdarnell


      cluster: Create final cluster version for 2.1

Release note: None

4b8c297


      importccl: ignore mysql's _binary introducer

MySQL version 5.7.23 included a change to mysqldump, compared to .22, to prefix binary fields with the '_binary' introducer.
These are described in the mysql docs here: https://dev.mysql.com/doc/refman/5.7/en/charset-introducer.html

Fixes #31169.

Release note (sql change): handle binary fields dumped by muysqldump v5.7.23 with _binary prefix.

f04c6b7


      importccl: add explicit default to mysql testdata timestamp

this makes the testdata work on mysql 8.0.2+, where the timestamp type no longer has the implicit defaults.

Release note: none.

9e5eb46


      importccl: parse zero mysql dates as null

these nonsense values are permitted by older / less strict mysql
versions but have no cockroach counterpart -- a date is either null, or
an actual date, which cannot have a zero day or month. Mapping these
values to null is the closest we can get them, and in schemas that allow
null values, hopefully makes migration easier. In schemas that do not
allow nulls these will of course produce errors but we do not really
have another option in those cases.

Fixes #29298.

Release note: none.

73c2e31

craig[bot] and bdarnell


      Merge #31304

31304: backport-2.1: cluster: Create final cluster version for 2.1 r=a-robinson a=bdarnell

Release note: None

Co-authored-by: Ben Darnell <ben@bendarnell.com>

knz


      deps: bump go-libedit

Release note (build change): CockroachDB can now be built from source
on macOS 10.14 (Mojave).

4b5441a

knz


      Merge pull request #31299 from knz/backport2.1-31082

release-2.1: Update the `unique_rowid` description

9cf327f


      Merge pull request #31305 from dt/backport2.1-31188-31217-31288

release-2.1: mysql import polish fixes

8c5fd36

petermattis


      ccl/storageccl/engineccl: properly handle intents which straddle ssta…

…bles

An intent which straddles an sstable can lead an incremental iterator to
incorrectly ignore an sstable. In order to fix this, when an intent
straddles an sstable (i.e. the metadata key is the last key in the
sstable) we need to include the intent's timestamp in the timestamp
bounds. We don't need to do this for interior intents because we'll
already be including the intent's timestamp as it is contained in the
next key following the intent. Add
`TestMVCCIncrementalIteratorIntentStraddlesSStables` which demonstrates
the problem.

Fixes #28358

Release note (bug fix): Fix a rare scenario where a backup could
incorrectly include a key for a transaction which was aborted.

dbb839a

petermattis


      Merge pull request #31316 from petermattis/backport2.1-31290

release-2.1: ccl/storageccl/engineccl: properly handle intents which straddle sstables

051be3c

changangela


      Merge pull request #31283 from changangela/20181011-backport-31216

release-2.1: sql: add merge joiner benchmark and memory account shrinking optimiza…

f94b4b6

tschottdorf


      Merge pull request #31251 from tschottdorf/backport2.1-30919

backport-2.1: storage: de-flake TestStoreRangeWaitForApplication

db85d17

tschottdorf


      roachtest: print output on failure in election test

This was hiding the output if the invocation itself failed, which is
when you wanted it most.

Release note: None

629267e

tschottdorf


      kv: improve a trace event about descriptor eviction

Release note: None

4d0fb44

tschottdorf


      storage: report optional StoreID with RangeNotFoundError

Release note: None

9998c7a

tschottdorf


      kv: try next replica on RangeNotFoundError

Previously, if a Batch RPC came back with a RangeNotFoundError,
we would immediately stop trying to send to more replicas, evict the
range descriptor, and start a new attempt after a back-off.

This new attempt could end up using the same replica, so if the
RangeNotFoundError persisted for some amount of time, so would the
unsuccessful retries for requests to it as DistSender doesn't
aggressively shuffle the replicas.

It turns out that there are such situations, and the
election-after-restart roachtest spuriously hit one of them:

1. new replica receives a preemptive snapshot and the ConfChange
2. cluster restarts
3. now the new replica is in this state until the range wakes
   up, which may not happen for some time.
4. the first request to the range runs into the above problem

@nvanbenschoten: I think there is an issue to be filed about the
tendency of DistSender to get stuck in unfortunate configurations.

Fixes #30613.

Release note (bug fix): Avoid repeatedly trying a replica that was found
to be in the process of being added.

6aef2e4

tschottdorf


      kv: cache leaseholder on successful response

Whenever a successful response is received from an RPC that we know has
to contact the leaseholder to succeed, update the leaseholder cache.

The immediate motivation for this is to be able to land the preceding
commits, which greatly exacerbated (as in, added a much faster failure
mode to)

```
make stress PKG=./pkg/sql/logictest TESTS=TestPlannerLogic/5node-dist/distsql_interleaved_join
```

However, the change is one we've wanted to make for a while; our caching
and in particular the eviction of leaseholders has been deficient
essentially ever since it was first introduced.

Touches #31068.

Release note: None

fdca0bd

tschottdorf


      Merge pull request #31250 from tschottdorf/backport2.1-31013

backport-2.1: kv: try next replica on RangeNotFoundError

a7a3cc1

Commits on Oct 13, 2018

knz


      Merge pull request #31308 from knz/20181012-bump-libedit2

release-2.1: deps: bump go-libedit

c6f482a

knz


      Merge pull request #31280 from knz/backport2.1-31222

release-2.1: sql: assorted mutation fixes

a44d231

Commits on Oct 14, 2018

asubiotto


      Merge pull request #31278 from asubiotto/backport2.1-31277

release-2.1: sql: make DISTSQL option to EXPLAIN ANALYZE optional

3fd21f7

Commits on Oct 15, 2018

knz


      sql: make heuristic planner bail on HAVING without FROM

The heuristic planner is unable to plan HAVING without FROM
properly. Instead of silently and incorrectly accepting this
construct, this patch makes it error out.

Release note (bug fix): CockroachDB does not any more silently and
incorrectly ignore the HAVING clause on SELECT without FROM.

1fe3088

nvanbenschoten


      sql: bump SET CLUSTER SETTING gossip timeout

This change bumps the timeout that SET CLUSTER SETTING statements
wait for gossip to reflect their expected value. I had seen this
timeout fail when running in highly-distributed clusters under load.
10 seconds seems more reasonable.

Release note: None

b26a148

knz


      Merge pull request #31300 from knz/backport2.1-29653

release-2.1: cli: minimize --certs-dir flag description.

b0b2054

benesch


      storage: eagerly queue undersize ranges for merging

When processing a write, check if the range has dropped beneath the
minimum size threshold. If it has, check whether the merge queue would
like to process it. This is analogous to our logic for eagerly splitting
oversize ranges.

Fix #29004.

Release note: None

58ad267

knz


      Merge pull request #31347 from knz/backport2.1-31346

release-2.1: sql: make heuristic planner bail on HAVING without FROM

f4ec893

knz


      sql/parser: annotate unimplemented features with suitable errors

This patch recognizes more syntax from PostgreSQL for the purpose of
reporting more fine grained "unimplemented feature" errors.

- new:
  - CREATE/DROP
    AGGREGATE / CAST / COLLATION / CONVERSION / DOMAIN / EXTENSION /
    FOREIGN TABLE / FOREIGN DATA (WRAPPER) / FUNCTION / LANGUAGE /
    OPERATOR / PUBLICATION / RULE / SERVER / SUBSCRIPTION /
    TEXT (SEARCH) / TRIGGER / MATERIALIZED VIEW.
  - CREATE OR REPLACE VIEW
  - CREATE RECURSIVE VIEW
  - CREATE TEMP TABLE / VIEW / SEQUENCE and variants
  - CREATE INDEX USING GIST / HASH / SPGIST / BRIN
  - CREATE INDEX ... WHERE ... (partial indexes)
  - CREATE INDEX ON ... (<complex expr>)   (computed indexes)
  - CREATE GROUP is really an alias for CREATE USER and is not "unsupported"
  - type names point/circle/etc   (unsupported pg data types)
  - INTERVAL SECONDS / MINUTES / etc  (interval type with unit qualifier)
  - SELECT .. FOR UPDATE (and other locks)
  - Geometric types: POINT, CIRCLE etc
  - Other predefined pg types: XML, MACADDR, etc
  - Full text search: CREATE TEXT SEARCH, types TSVECTOR, TSQUERY, etc

- modified:
  - COMMENT ON accounting split in "table", "column" and other
  - TIMETZ support accounting split in "data type" and "current_time"
  - WITH RECURSIVE gets an issue number

Additionally unimplemented feature errors that stem during parsing
will now be prefixed with `syntax.` to ease integration in the feature
count hierarchy.

Release note: None

2d1751d

knz


      telemetry: minor optimization

Release note: None

cbc7ff6

solongordon


      distsql: fix nil pointer in sort benchmarks

Sorter benchmarks were failing with nil pointer errors due to a missing
disk monitor.

Release note: None

d4dc202

knz


      sql: fold unimplemented/error telemetry into feature counters

This was discussed with @dt: use the feature counter machinery
instead of using separate maps. This simplifies the code.

Release note: None

101dd70

knz


      sql: report unimplemented vtables in telemetry

Release note (sql change): CockroachDB will now collect references to
tables in `information_schema` and `pg_catalog` which are not yet
implemented, and report them as telemetry if statistics reporting is
enabled. This will help determine which features should be implemented
next for compatibility.

66f3576

knz


      pgerror: hint what to do upon encountering unimplemented errors

Before:

```
root@127.0.0.1:28070/defaultdb> select * from information_schema.view_table_usage;
pq: virtual schema table not implemented: information_schema.view_table_usage
```

After:

```
root@127.0.0.1:28070/defaultdb> select * from information_schema.view_table_usage;
pq: virtual schema table not implemented: information_schema.view_table_usage
HINT: This feature is not yet implemented in CockroachDB.

Please check https://github.com/cockroachdb/cockroach/issues to check
whether this feature is already tracked. If you cannot find it there,
please report this error with example query at:

    https://github.com/cockroachdb/cockroach/issues/new/choose

If you would rather not post publicly, please contact us directly at:

    support@cockroachlabs.com

The Cockroach Labs team appreciates your feedback.
```

Release note: None

5c15f48

benesch


      Merge pull request #31352 from benesch/backport2.1-30590

release-2.1: storage: eagerly queue undersize ranges for merging

f0fb532

benesch


      build: check out submodules before yarn install

`yarn install` requires the yarn-vendor submodule to be available.
This was missed when we started vendoring our Yarn dependencies in
PR #27035.

Fix #30619.

Release note: None

b649df3

benesch


      storage: avoid merge deadlock when concurrent split wins

When merging two adjacent ranges P and Q, we need to be careful to
delete Q's range descriptor with a conditional put, or we can deadlock
with a concurrent split. See the comments and test case within for
details. This commit fixes the last issue that is blocking merges from
being turned on by default.

Release note: None

3e41de6

BramGruneir


      ui: Have the range debug page correctly handle missing lease times

Before this change, the value was always assumed to be not-null and it was null
would crash. It will now correctly handled the missing value and display a
`no timestamp` warning.

Fixes #31260.

Release note (bug fix): The range debug page will now correctly handle cases in
which there is no lease start or expiration time.

7d8e24b

nvanbenschoten


      Merge pull request #31348 from nvanbenschoten/backport2.1-31194

release-2.1: sql: bump SET CLUSTER SETTING gossip timeout

b0ca488

benesch


      Merge pull request #31365 from benesch/backport2.1-30993

release-2.1: build: check out submodules before yarn install

f29847d

benesch


      Merge pull request #31366 from benesch/backport2.1-31294

release-2.1: storage: avoid merge deadlock when concurrent split wins

8c8cdce

BramGruneir


      Merge pull request #31367 from BramGruneir/backport2.1-31362

release-2.1: ui: Have the range debug page correctly handle missing lease times

b379f28

a-robinson


      storage: Avoid using different qps measurements in StoreRebalancer

It's better to use the same measurement for everything, especially
because it's the one that was used to compute the store's total QPS.

Fixes #31303

Release note: None

abb4873

a-robinson


      storage: Jitter the StoreRebalancer loop's timing

Release note: None

cb1ac4e


      Merge pull request #31357 from knz/backport2.1-31335-31356-31332-31333

release-2.1: assorted SQL telemetry fixes

429569b

a-robinson


      Merge pull request #31379 from a-robinson/backport2.1-31324

release-2.1: storage: Avoid using different qps measurements in StoreRebalancer

ddbc2b8

a-robinson


      Merge pull request #31381 from a-robinson/backport2.1-31227

release-2.1: storage: Jitter the StoreRebalancer loop's timing

c0bc4e0

benesch


      testcluster: fully disable the merge queue in manual replication mode

The merge queue needs to be disabled via cluster setting (and not
testing knob) in order for ALTER TABLE ... SPLIT AT to work.

Release note: None

08c17df

danhhz


      changefeedccl: add a test for many changefeeds watching the same table

Closes #31197

Release note: None

a6bc879

danhhz


      changefeedccl: test changefeed with unspecified primary key

If the user doesn't specify a primary key when making the table, we add
one (a hidden `unique_rowid` column). This tests the behavior of a
changefeed over one of these tables.

Closes #31202

Release note: None

ffe308f

danhhz


      changefeedccl: test cursor= edge cases

Closes #29661

Release note: None

d62c1de

solongordon


      Merge pull request #31360 from solongordon/backport2.1-31208

release-2.1: distsql: fix nil pointer in sort benchmarks

21dd8ed

benesch


      storage: deflake TestLogMerges

This test needs the merge queue disabled.

Fix #31061.

Release note: None

c7c4e2f

benesch


      storage: deflake TestStoreSetRangesMaxBytes

TestStoreSetRangesMaxBytes creates manual splits (using a horrendously
outdated function, but that's a problem for another day), so it can't
tolerate the merge queue.

Release note: None

46535e8

benesch


      distsqlplan: deflake TestMixedDirections and TestSpanResolver

These tests both need the merge queue to be disabled.

Fix #29321.
Fix #31060.

Release note: None

083e4b3

benesch


      Merge pull request #31394 from benesch/backport2.1-29583

release-2.1: storage: re-enable the merge queue by default

062e5cd

Commits on Oct 16, 2018

vilterp


      server(ccl)/ui: identify zone configs by zone name, not ID

Previously, the DataDistribution endpoint was returning a map of zone
configs by zone config ID. However, zone config ID is not a unique
identier within `crdb_internal.zones`, since subzones attached to
partitions share the same ID (that of the top-level zone config).

Zone names should be unique; this change uses those instead and updates
the UI accordingly.

This change also introduces a `serverccl` package for the unit test to
live in, since testing this code path requires creating partitions, a
CCL feature.

Release note: None

d7254ab

benesch


      storage: actually enable the merge queue

This got lost in the backport (#31394) due to a merge conflict.

Release note: None

3e69f3a

RaduBerinde


      Merge pull request #31419 from benesch/merge-enable

storage: actually enable the merge queue

e6348bb

danhhz


      Merge pull request #31397 from danhhz/backport2.1-31392

release-2.1: changefeedccl: more tests

464baec

vilterp


      Merge pull request #31417 from vilterp/backport2.1-31407

release-2.1: server(ccl)/ui: identify zone configs by zone name, not ID

4810e59

benesch


      c-deps: bump CryptoPP to avoid SIGTRAP on macOS

Bump CryptoPP to pick up a fix for #31380.
Details reproduced below.

Fix #31380.

---

As part of its CPU feature detection, CryptoPP installs a SIGILL signal
handler before issuing the cpuid instruction. The intent is to
gracefully degrade on CPUs that don't support the cpuid instruction.

The problem is that it is impossible to safely overwrite a signal
handler installed by the Go runtime in go1.10 on macOS
(golang/go#22805). This causes CockroachDB 2.0 to crash on macOS Mojave:
#31380.

The situation has improved on the Go front, as go1.11 makes it possible
to safely save and restore signal handlers installed by the Go runtime
on macOS.

Still, we can do better and support go1.10. There is no need to bother
installing a SIGILL handler, as the cpuid instruction is supported by
every x86-64 CPU. We can instead use conditional compilation to make
sure that we never execute a cpuid instruction on a non x86-64 CPU.

Note that CPU feature detection is performed at executable load time
(see the attribute(constructor) on DetectX86Features); therefore any
reference to function which calls DetectX86Features (notably HasAESNI)
corrupts the signal handler. It's not entirely clear why this corruption
later leads to the SIGTRAP seen in #31380--is
something in macOS or the Go runtime generating a SIGILL and trying to
handle it gracefully?--but regardless, not mucking with the signal
handler fixes the issue.

Release note (bug fix): CockroachDB no longer crashes due to a SIGTRAP error
soon after startup on macOS Mojave (#31380).

4d87813

benesch


      Merge pull request #31520 from benesch/backport2.1-31516

release-2.1: c-deps: bump CryptoPP to avoid SIGTRAP on macOS

670c7d3

Commits on Oct 17, 2018

mrtracy


      cdc: Fix Sink Retries Again

Another small error made in the implementation of sink retries - in
order to provide a good UX for the CREATE CHANGEFEED statement, we want
to immediately service sink-related errors to the user when the
statement is executing. Once the changefeed job is created, however, we
retry (and log) errors from the sink. In the previous implementation,
this check was performed when the changefeed distsql processors were
instantiated; however, this occurs inside of the "Resume" method, which
can not easily tell if there is still a user statement waiting for
feedback. Therefore, we were not retrying on some errors that we should
have been retrying.

With this change, the "canary" sink used to detect these errors is moved
to the statement hook, which is only executed as part of the SQL
statement and not as part of the job's Resume loop, and therefore is not
subject to retries.

Release note: Fixed a bug where changefeeds may not correctly retry
temporary errors when communicating with a sink.

af6eb4b

mrtracy


      Merge pull request #31559 from mrtracy/backport2.1-31410

backport-2.1: cdc: Fix Sink Retries Again

e479b7a

Unified Split

Showing 1,528 changed files with 89,659 additions and 35,696 deletions.

Show comments View

23 .github/ISSUE_TEMPLATE.md

This file was deleted.

Show comments View

46 .github/ISSUE_TEMPLATE/bug_report.md

          
    @@ -0,0 +1,46 @@
  
    ---

    name: Bug or crash report

    about: Report unexpected behavior to help us improve

    ---

    **Describe the problem**

    Please describe the issue you observed, and any steps we can take to reproduce it:

    **To Reproduce**

    What did you do? Describe in your own words.

    If possible, provide steps to reproduce the behavior:

    1. Set up CockroachDB cluster ...

    2. Send SQL ... / CLI command ...

    3. Look at UI / log file / client app ...

    4. See error

    **Expected behavior**

    A clear and concise description of what you expected to happen.

    **Additional data / screenshots**

    If the problem is SQL-related, include a copy of the SQL query and the schema

    of the supporting tables.

    If a node in your cluster encountered a fatal error, supply the contents of the

    log directories (at minimum of the affected node(s), but preferably all nodes).

    Note that log files can contain confidential information. Please continue

    creating this issue, but contact support@cockroachlabs.com to submit the log

    files in private.

    If applicable, add screenshots to help explain your problem.

    **Environment:**

     - CockroachDB version [e.g. 2.0.x]

     - Server OS: [e.g. Linux/Distrib]

     - Client app [e.g. `cockroach sql`, JDBC, ...]

    **Additional context**

    What was the impact?

    Add any other context about the problem here.

Show comments View

17 .github/ISSUE_TEMPLATE/feature_request.md

          
    @@ -0,0 +1,17 @@
  
    ---

    name: Feature request

    about: Suggest an idea for this project

    ---

    **Is your feature request related to a problem? Please describe.**

    A clear and concise description of what the problem is. Ex. I'm always frustrated when [...]

    **Describe the solution you'd like**

    A clear and concise description of what you want to happen.

    **Describe alternatives you've considered**

    A clear and concise description of any alternative solutions or features you've considered.

    **Additional context**

    Add any other context or screenshots about the feature request here.

Show comments View

60 .github/ISSUE_TEMPLATE/performance-inquiry.md

          
    @@ -0,0 +1,60 @@
  
    ---

    name: Performance inquiry

    about: You have a question about CockroachDB's performance and it is not a bug or

      a feature request

    ---

    **What is your situation?**

    Select all that apply:

    - is there a difference between the performance you expect and the performance you observe?

    - do you want to improve the performance of your app?

    - are you surprised by your performance results?

    - are you comparing CockroachDB with some other database?

    - another situation? Please explain.

    **Observed performance**

    What did you see? How did you measure it?

    If you have already ran tests, include your test details here:

    - which test code do you use?

    - which SQL queries? Schema of supporting tables?

    - how many clients per node?

    - how many requests per client / per node?

    **Application profile**

    Performance depends on the application. Please help us understand how you use CockroachDB before we can discuss performance.

    - Have you used other databases before? Or are you considering a migration? Please list your previous/other databases here.

    - What is the scale of the application?

      - how many columns per table?

      - how many rows (approx) per table?

      - how much data?

      - how many clients? Requests / second?

    - What is the query profile?

      - is this more a OLTP/CRUD workload? Or Analytics/OLAP? Is this hybrid/HTAP?

      - what is the ratio of reads to writes?

      - which queries are grouped together in transactions?

    - What is the storage profile?

      - how many nodes?

      - how much storage?

      - how much data?

      - replication factor?

    **Requested resolution**

    When/how would you consider this issue resolved? 

    Select all that applies:

    - I mostly seek information: data, general advice, clarification.

    - I seek guidance as to how to tune my application or CockroachDB deployment.

    - I want CockroachDB to be optimized for my use case.

cockroachdb/cockroach

Comparing changes

Open a pull request

No commit comments for this range