Skip to content

Fix revivification CLI flag ordering dependency#1644

Merged
vazois merged 3 commits intomicrosoft:mainfrom
hamdaankhalid:garnet-hkhalid-fix-cli-ordering
Mar 27, 2026
Merged

Fix revivification CLI flag ordering dependency#1644
vazois merged 3 commits intomicrosoft:mainfrom
hamdaankhalid:garnet-hkhalid-fix-cli-ordering

Conversation

@hamdaankhalid
Copy link
Copy Markdown
Contributor

The validation in Options.GetServerOptions() checked the raw enableRevivification flag (from --reviv) instead of the computed useRevivBinsPowerOf2 state. This caused specifying --reviv alongside explicit --reviv-bin-record-sizes and --reviv-bin-record-counts to always fail, despite the --reviv help text documenting that it can be overridden by the combination of these flags.

Changed the check from enableRevivification to useRevivBinsPowerOf2, which correctly reflects that --reviv-bin-record-sizes already overrides the power-of-2 default when present.

The validation in Options.GetServerOptions() checked the raw
enableRevivification flag (from --reviv) instead of the computed
useRevivBinsPowerOf2 state. This caused specifying --reviv alongside
explicit --reviv-bin-record-sizes and --reviv-bin-record-counts to
always fail, despite the --reviv help text documenting that it can be
overridden by the combination of these flags.

Changed the check from enableRevivification to useRevivBinsPowerOf2,
which correctly reflects that --reviv-bin-record-sizes already
overrides the power-of-2 default when present.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Copilot AI review requested due to automatic review settings March 24, 2026 18:38
Copy link
Copy Markdown
Contributor

Copilot AI left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull request overview

Fixes revivification CLI flag validation so that --reviv can be combined with explicit --reviv-bin-record-sizes / --reviv-bin-record-counts regardless of argument ordering, matching the documented override behavior.

Changes:

  • Update Options.GetServerOptions() validation to check useRevivBinsPowerOf2 (computed state) instead of the raw enableRevivification flag.
  • Add a new config-parsing test covering multiple argument orderings for --reviv with explicit bin sizes/counts, plus sizes-only and counts-only scenarios.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File Description
test/Garnet.test/GarnetServerConfigTests.cs Adds coverage ensuring revivification-related CLI flags parse consistently across multiple argument orderings.
libs/host/Configuration/Options.cs Adjusts revivification option validation to reflect overridden “power-of-2” behavior when custom bins are provided.
Comments suppressed due to low confidence (1)

libs/host/Configuration/Options.cs:792

  • --reviv-bin-record-sizes without --reviv-bin-record-counts is intended to work (defaulting each bin to 256 records per help text/docs). However, RevivBinRecordCounts comes from defaults.conf as null, so when only sizes are provided hasRecordCounts is false and this block is skipped, leaving RevivBinRecordCounts potentially null in GarnetServerOptions. Later, the revivification bin setup code uses RevivBinRecordCounts?.Length switch { 0 => ..., 1 => ..., _ => RevivBinRecordCounts[ii] }, which will hit the _ arm when counts is null and can throw at runtime. Consider normalizing revivBinRecordCounts to an empty array when sizes are specified but counts are not, or updating the downstream switch to treat null the same as length 0.
            if (hasRecordCounts)
            {
                if (useRevivBinsPowerOf2)
                    throw new Exception("Revivification cannot specify both record counts and powerof2 bins.");
                if (!hasRecordSizes)
                    throw new Exception("Revivification bin counts require bin sizes.");
                useRevivBinsPowerOf2 = false;

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

@vazois vazois merged commit 3610cd6 into microsoft:main Mar 27, 2026
82 of 93 checks passed
vazois added a commit that referenced this pull request Mar 27, 2026
The validation in Options.GetServerOptions() checked the raw
enableRevivification flag (from --reviv) instead of the computed
useRevivBinsPowerOf2 state. This caused specifying --reviv alongside
explicit --reviv-bin-record-sizes and --reviv-bin-record-counts to
always fail, despite the --reviv help text documenting that it can be
overridden by the combination of these flags.

Changed the check from enableRevivification to useRevivBinsPowerOf2,
which correctly reflects that --reviv-bin-record-sizes already
overrides the power-of-2 default when present.

Co-authored-by: Hamdaan Khalid <hkhalid@microsoft.com>
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Co-authored-by: Vasileios Zois <96085550+vazois@users.noreply.github.com>
vazois added a commit that referenced this pull request Apr 2, 2026
* Add es-metadata.yml (#1641)

* Fix INFO all/default/everything returning empty response (#1645)

* Fix INFO all/default/everything returning empty response

Implement support for INFO all, default, and everything options:
- all: returns all DefaultInfo sections excluding module-generated ones
- default: returns the default set of sections (same as no-arg INFO)
- everything: returns all DefaultInfo sections including modules

Added pre-declared HashSet collections (AllInfoSet, EverythingInfoSet)
in GarnetInfoMetrics.cs derived from DefaultInfo to support these options.

Updated InfoCommand.cs to use UnionWith with the new HashSets instead of
silently skipping the ALL keyword.

Added tests for all three options, verifying correct section inclusion/exclusion.

Fixes #1643

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* make everything option use DefaultInfo

* Add null/empty guard in GetSectionHeaders test helper

Adds explicit asserts before splitting INFO output so test failures
surface a clear message instead of a NullReferenceException.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* wip; restructuring cluster tests to reduce CI duration

* separate dispose from close and configure socket to allow rapid connect

* ensure socket is disposed succesfully

* fix failing test

* update global.json

---------

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Fix revivification CLI flag ordering dependency (#1644)

The validation in Options.GetServerOptions() checked the raw
enableRevivification flag (from --reviv) instead of the computed
useRevivBinsPowerOf2 state. This caused specifying --reviv alongside
explicit --reviv-bin-record-sizes and --reviv-bin-record-counts to
always fail, despite the --reviv help text documenting that it can be
overridden by the combination of these flags.

Changed the check from enableRevivification to useRevivBinsPowerOf2,
which correctly reflects that --reviv-bin-record-sizes already
overrides the power-of-2 default when present.

Co-authored-by: Hamdaan Khalid <hkhalid@microsoft.com>
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Co-authored-by: Vasileios Zois <96085550+vazois@users.noreply.github.com>

* [Bugfix] SortedSet Race Condition - Mutating under Shared Lock (#1642)

* fixed

* SortedSet Race Condition - Mutation under Shared Lock

---------

Co-authored-by: Hamdaan Khalid <hkhalid@microsoft.com>
Co-authored-by: Vasileios Zois <96085550+vazois@users.noreply.github.com>

* Bugfix: Accept Loop socket error handling (#1646)

* Bugfix Accept Loop

* race condition and tiny loggin fix

* Dispose potentially partial sockets.

* Fix accept loop zombie state: tiered error handling with backoff retry

The accept loop in GarnetServerTcp.HandleNewConnection previously treated
all SocketError != Success as fatal, disposing the SocketAsyncEventArgs and
permanently killing the accept loop. This left the server in a zombie state
where existing connections worked but no new connections were accepted.

Changes:
- Extract HandleAcceptError with three-tier error categorization:
  - Tier 1 (fatal): OperationAborted, NotSocket, etc. — stop loop
  - Tier 2 (resource pressure): TooManyOpenSockets, NoBufferSpaceAvailable,
    etc. — backoff via Timer and retry (100ms initial, 5s cap)
  - Tier 3 (client-caused transient): ConnectionReset, etc. — log and continue
- Add ScheduleAcceptRetry using Timer to avoid blocking IOCP threads
- Dispose retry timer on server shutdown
- Add AcceptLoopTests with RST flood and extended attack simulation

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* PR comments

* Nit

* Nit

* simplify greatly

* easy win

* small updates

---------

Co-authored-by: Hamdaan Khalid <hkhalid@microsoft.com>
Co-authored-by: Badrish Chandramouli <badrishc@microsoft.com>
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Co-authored-by: Vasileios Zois <96085550+vazois@users.noreply.github.com>

* Skip Value Expiration Check When Scanning a Tombstoned Record (#1612)

* fix migrate write test

* prevent expiration check on tombstoned key while scanning

* fix formatting

* ensure reviv pause signal is observed through epoch protection

* make revivPauseEvent readonly

* Update test/Garnet.test.cluster/ClusterMigrateTests.cs

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

* addressing first comments

* update comment in MigrateScanFunctions

* release epoch when acquiring exlucisve SuspendConfigMerge lock

* add ReaderWriterLock custom implementation

* fix formatting

* fixing simple tests

* make pause reviv thread safe

---------

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

* Bump picomatch from 2.3.1 to 2.3.2 in /website (#1649)

Bumps [picomatch](https://github.com/micromatch/picomatch) from 2.3.1 to 2.3.2.
- [Release notes](https://github.com/micromatch/picomatch/releases)
- [Changelog](https://github.com/micromatch/picomatch/blob/master/CHANGELOG.md)
- [Commits](micromatch/picomatch@2.3.1...2.3.2)

---
updated-dependencies:
- dependency-name: picomatch
  dependency-version: 2.3.2
  dependency-type: indirect
...

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>

* Bump brace-expansion from 1.1.12 to 1.1.13 in /website (#1651)

Bumps [brace-expansion](https://github.com/juliangruber/brace-expansion) from 1.1.12 to 1.1.13.
- [Release notes](https://github.com/juliangruber/brace-expansion/releases)
- [Commits](juliangruber/brace-expansion@v1.1.12...v1.1.13)

---
updated-dependencies:
- dependency-name: brace-expansion
  dependency-version: 1.1.13
  dependency-type: indirect
...

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>

* Handle SocketException on accepted socket configuration (#1648)

When a client RSTs between accept completing and socket setup (e.g. NoDelay),
the SocketException was unhandled and crashed the process. Wrap the post-accept
socket configuration in a try/catch, dispose the dead socket, and continue
the accept loop.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Fix CLUSTER NODES and CLUSTER SHARDS metadata output (#1657)

* Fix CLUSTER NODES and CLUSTER SHARDS metadata output (#1650)

- CLUSTER NODES: Only append ,hostname when hostname is non-empty
- CLUSTER SHARDS: Replace 'address' field with Redis-compatible 'ip',
  'endpoint', and optional 'hostname' fields
- CLUSTER SHARDS: Honor ClusterPreferredEndpointType for endpoint field
  (ip address when Ip, hostname when Hostname)
- CLUSTER SHARDS: Fix role output to use 'master'/'slave' instead of
  enum names 'PRIMARY'/'REPLICA'
- Update NodeInfo struct and ClusterShards parser in test utilities
  for dynamic key-value field parsing
- Add ClusterShardsTest and ClusterNodesHostnameTest unit tests

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Refactor CLUSTER SLOTS/SHARDS to use StringBuilder

Replace string concatenation with StringBuilder in GetShardsInfo,
GetSlotsInfo, and their helper methods (AppendFormattedSlotInfo,
AppendNodeNetworkingInfo) to reduce intermediate string allocations.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Address PR review comments and fix endpoint validation

- Fix endpoint validation (Options.cs): allow cluster-announce-ip when
  bind address is 0.0.0.0/:: (wildcard), since the server listens on
  all interfaces. Only require port match in that case.
- Handle ClusterPreferredEndpointType.Unknown in CLUSTER SHARDS: set
  endpoint to '?' consistent with CLUSTER SLOTS and redirects.
- Add ClusterShardsTest cases for Unknown endpoint type.
- Improve ClusterNodesHostnameTest to handle both hostname-present
  and hostname-absent branches.
- Add even-length guard and explicit role validation in ClusterShards
  test parser.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

---------

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Apply suggestions from code review

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

---------

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: Badrish Chandramouli <badrishc@microsoft.com>
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Co-authored-by: Hamdaan Khalid <42720645+hamdaankhalid@users.noreply.github.com>
Co-authored-by: Hamdaan Khalid <hkhalid@microsoft.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants