Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

DNM: backport stretch clusters to nautilus #37173

Closed
wants to merge 167 commits into from
Closed

DNM: backport stretch clusters to nautilus #37173

wants to merge 167 commits into from

Conversation

gregsfortytwo
Copy link
Member

I don't think we want to merge this, but I've done the work for downstream. Putting up as an FYI. (Hopefully I can find the draft mode thing, but it's not visible yet?)

#35906

Checklist

  • References tracker ticket
  • Updates documentation if necessary
  • Includes tests for new functionality or reproducer for bug
Show available Jenkins commands
  • jenkins retest this please
  • jenkins test classic perf
  • jenkins test crimson perf
  • jenkins test signed
  • jenkins test make check
  • jenkins test make check arm64
  • jenkins test submodules
  • jenkins test dashboard
  • jenkins test api
  • jenkins test docs
  • jenkins render docs
  • jenkins test ceph-volume all
  • jenkins test ceph-volume tox

@gregsfortytwo
mon: Rearrange Paxos::dispatch to be a little cleaner
8c03507 
The Message is actually always a MMonPaxos (which is not a subtype of
PaxosServiceMessage, so that whole thing was wrong!). This lets
us clean up the switch quite a bit and correct the odd casts
going around.

Signed-off-by: Greg Farnum <gfarnum@redhat.com>
(cherry picked from commit 0c85f8f)
Resolves: rhbz#1800382
@gregsfortytwo
elector: hoist Elector data bits out into a new ElectionLogic class
77a11ff 
Signed-off-by: Greg Farnum <gfarnum@redhat.com>
(cherry picked from commit cda3b25)

Conflicts:
  src/mon/Elector.cc

Signed-off-by: Greg Farnum <gfarnum@redhat.com>
Resolves: rhbz#1800382
@gregsfortytwo
elector: store accessors for ElectionLogic
109d5bc 
Signed-off-by: Greg Farnum <gfarnum@redhat.com>
(cherry picked from commit 987630f)
Resolves: rhbz#1800382
@gregsfortytwo
elector: hoist bump_epoch into ElectionLogic
58c9a8a 
Mangle some pointers and the dout prefix to make things compile
during this transition.

Signed-off-by: Greg Farnum <gfarnum@redhat.com>
(cherry picked from commit 2069016)

Resolves: rhbz#1800382
@gregsfortytwo
elector: split handle_propose in two and hoist into ElectionLogic
c667e15 
Also add a bunch of utility functions that will need to
get better implementations later

Signed-off-by: Greg Farnum <gfarnum@redhat.com>
(cherry picked from commit e24a73d)

Resolves: rhbz#1800382
@gregsfortytwo
elector: hoist defer into ElectionLogic
9ead769 
Signed-off-by: Greg Farnum <gfarnum@redhat.com>
(cherry picked from commit a227a24)

Resolves: rhbz#1800382
@gregsfortytwo
elector: hoist init into ElectionLogic
ed271f3 
Signed-off-by: Greg Farnum <gfarnum@redhat.com>
(cherry picked from commit 0205695)

Resolves: rhbz#1800382
@gregsfortytwo
elector: hoist participating into ElectionLogic
efdd2a9 
Signed-off-by: Greg Farnum <gfarnum@redhat.com>
(cherry picked from commit b53b984)

Resolves: rhbz#1800382
@gregsfortytwo
elector: hoist start into ElectionLogic
7ea003c 
Signed-off-by: Greg Farnum <gfarnum@redhat.com>
(cherry picked from commit 952d785)

Resolves: rhbz#1800382
@gregsfortytwo
elector: hoist expire into ElectionLogic
9d2ac8a 
Signed-off-by: Greg Farnum <gfarnum@redhat.com>
(cherry picked from commit 3ca09c9)

Resolves: rhbz#1800382
@gregsfortytwo
elector: hoist victory into ElectionLogic
a9114d8 
Signed-off-by: Greg Farnum <gfarnum@redhat.com>
(cherry picked from commit 469910b)

Conflicts:
  src/mon/Elector.cc

Signed-off-by: Greg Farnum <gfarnum@redhat.com>
Resolves: rhbz#1800382
@gregsfortytwo
elector: hoist handle_ack into ElectionLogic
9048a1e 
Signed-off-by: Greg Farnum <gfarnum@redhat.com>
(cherry picked from commit 925f132)

Resolves: rhbz#1800382
@gregsfortytwo
elector: hoist handle_victory into ElectionLogic
150e2a1 
Signed-off-by: Greg Farnum <gfarnum@redhat.com>
(cherry picked from commit 1d1c143)

Resolves: rhbz#1800382
@gregsfortytwo
elector: rename handle_propose_logic -> receive_propose
ac5aca4 
Signed-off-by: Greg Farnum <gfarnum@redhat.com>
(cherry picked from commit ba37ca9)

Resolves: rhbz#1800382
@gregsfortytwo
elector: route _bump_epoch through the interface-to-be
53abf8b 
Signed-off-by: Greg Farnum <gfarnum@redhat.com>
(cherry picked from commit b7a44c0)

Resolves: rhbz#1800382
@gregsfortytwo
elector: make ElectionLogic::declare_victory private
3753561 
Signed-off-by: Greg Farnum <gfarnum@redhat.com>
(cherry picked from commit c591bd4)

Resolves: rhbz#1800382
@gregsfortytwo
elector: create declare_standlone_victory in Elector/Logic for Monitor
b43158b 
This lets us make a few more ElectionLogic functions private.

Signed-off-by: Greg Farnum <gfarnum@redhat.com>
(cherry picked from commit 40339cd)

Resolves: rhbz#1800382
@gregsfortytwo
elector: make ElectionLogic private to Elector; undo most public shen…
da9af1a 
…anigans

Signed-off-by: Greg Farnum <gfarnum@redhat.com>
(cherry picked from commit bf9523b)

Resolves: rhbz#1800382
@gregsfortytwo
elector: move all the elector callouts into the Elector
6e3d390 
In the next commit we'll actually use the ElectionOwner interface.

Signed-off-by: Greg Farnum <gfarnum@redhat.com>
(cherry picked from commit 8a95e2c)

Resolves: rhbz#1800382
@gregsfortytwo
elector: split ElectionLogic into its own compilation unit
ec3653c 
Just copy the files, then delete the inappropriate bits.
Switched to using ldout in ElectionLogic.cc, added a few
functions I missed to the ElectionOwner interface, and moved
Elector() into the .cc so I could grab cct out of it.

Signed-off-by: Greg Farnum <gfarnum@redhat.com>
(cherry picked from commit 8eac583)

Conflicts:
	src/mon/Elector.cc

Signed-off-by: Greg Farnum <gfarnum@redhat.com>
Resolves: rhbz#1800382
@gregsfortytwo
elector: don't clear peer_info in bump_epoch()
8dbd49f 
Not sure how I got that wrong, but we bump_epoch() on
winning an election, right before we start looking through
peer_info. So just do it on election start instead.

Signed-off-by: Greg Farnum <gfarnum@redhat.com>
(cherry picked from commit 1261c44)

Resolves: rhbz#1800382
@gregsfortytwo
elector: fix privacy and restore dout in Elector
f5fb8da 
Signed-off-by: Greg Farnum <gfarnum@redhat.com>
(cherry picked from commit 94c2fd3)

Resolves: rhbz#1800382
@gregsfortytwo
elector: make some more ElectionLogic members private
c0f7ed8 
Signed-off-by: Greg Farnum <gfarnum@redhat.com>
(cherry picked from commit 28729ea)

Resolves: rhbz#1800382
@gregsfortytwo
elector: persist the epoch in bump_epoch()
2747983 
I somehow broke this when initially hoisting it out of Elector. Whoops!

Signed-off-by: Greg Farnum <gfarnum@redhat.com>
(cherry picked from commit b8f04c5)

Resolves: rhbz#1800382
@gregsfortytwo
elector: Update Elector and ElectionLogic function documentation
5a70190 
The world has changed, and now so have the nice docs.

Signed-off-by: Greg Farnum <gfarnum@redhat.com>
(cherry picked from commit 521a622)

Resolves: rhbz#1800382
@gregsfortytwo
elector: swap order of parameters in ElectionLogic::receive_propose
1fa1c6a 
Signed-off-by: Greg Farnum <gfarnum@redhat.com>
(cherry picked from commit fba2a0d)

Resolves: rhbz#1800382
@gregsfortytwo
elector: const-ify a bunch of functions
77b6b2e 
Signed-off-by: Greg Farnum <gfarnum@redhat.com>
(cherry picked from commit a56f620)

Resolves: rhbz#1800382
@gregsfortytwo
test: add new ElectionLogic unit test framework
023ccfe 
Signed-off-by: Greg Farnum <gfarnum@redhat.com>
(cherry picked from commit 5a64cdf)

Resolves: rhbz#1800382
@gregsfortytwo
test: switch to ldout; let users specify mon debug level
bc7c523 
Signed-off-by: Greg Farnum <gfarnum@redhat.com>
(cherry picked from commit c9583d5)

Resolves: rhbz#1800382
@gregsfortytwo
mon: elector: return after triggering a new election
f221105 
When receiving an old propose, we were correctly triggering a new election
but not then returning out of receive_propose(), so we processed the
"should I defer" logic and perhaps sent out a deferal (in the current epoch!).

Fixes: https://tracker.ceph.com/issues/42977

Signed-off-by: Greg Farnum <gfarnum@redhat.com>
(cherry picked from commit 337d12b)

Resolves: rhbz#1800382
@gregsfortytwo
mon: notify_new_monmap in all the places we update our monmap
2bf3701 
We can adopt new monmaps while bootstrapping, or in election messages, in
addition to MonmapMonitor::update_from_paxos. Since we use the
notification to update our election strategy and such, we need to notify
from these locations as well!

Fixes: https://tracker.ceph.com/issues/47654

Signed-off-by: Greg Farnum <gfarnum@redhat.com>
(cherry picked from commit 1d5f4e8)

Resolves: rhbz#1939766
@gregsfortytwo
mon: if in stretch mode, don't allow new mons to join without a location
b5c8a38 
We blocked off the other routes to add location-less monitors, but if you
turn on a monitor with the right keys it can auto-join via the MMonJoin
functionality. Block that off!

Signed-off-by: Greg Farnum <gfarnum@redhat.com>
(cherry picked from commit 6ca6c4b)

Resolves: rhbz#1939766
@gregsfortytwo
mon: messages: Extend MMonJoin so it can provide a crush_location on …
0698781 
…join

This will let newly-created monitors auto-join on startup in stretch mode, by
providing the needed location.

Signed-off-by: Greg Farnum <gfarnum@redhat.com>
(cherry picked from commit 9c0e32d)

Resolves: rhbz#1939766
@gregsfortytwo
mon: let users specify a crush location on boot, and send it in MMonJoin
8eafe3b 
Go to some effort to look at our location in the monmap and update it the
same way we update names or IP addresses when necessary. Let users
pass in the location on the CLI via "--set-crush-location".

Signed-off-by: Greg Farnum <gfarnum@redhat.com>
(cherry picked from commit 48886aa)

Resolves: rhbz#1939766
@gregsfortytwo
osd: osdmap: do not assert target_v when encoding under stretch mode
c4c40a3 
As in dd63a3e for the OSDMap, this
caused crashes when encoding for kernel clients, and is unnecessary
for servers because they are separately gated.

I did a full audit of every instance of "assert" I added to the codebase
to make sure this is the very last one of these issues.

Signed-off-by: Greg Farnum <gfarnum@redhat.com>
(cherry picked from commit 589de8b)

Resolves: rhbz#1944611
@gregsfortytwo
Copy link
Member Author

jenkins test make check

1 similar comment
@gregsfortytwo
Copy link
Member Author

jenkins test make check

@tchaikov
Copy link
Contributor

tchaikov commented Apr 1, 2021

the test failure should be fixed by #40536

@tchaikov
Copy link
Contributor

tchaikov commented Apr 1, 2021

jenkins test api

@gregsfortytwo
mon: rename maybe_engage_stretch_mode to try_engage_stretch_mode
e677192 
Signed-off-by: Greg Farnum <gfarnum@redhat.com>
(cherry picked from commit 596d09a)

Resolves: rhbz#1949166
@gregsfortytwo
mon: add a set_recovery_stretch_mode function
943f4f9 
Signed-off-by: Greg Farnum <gfarnum@redhat.com>
(cherry picked from commit e1ae925)

Resolves: rhbz#1949166
@gregsfortytwo
mon: maintain stretch_recovery_triggered in new OSDMon::set_*_stretch…
ba96727 
…_mode

Signed-off-by: Greg Farnum <gfarnum@redhat.com>
(cherry picked from commit 73b461a)

Resolves: rhbz#1949166
@gregsfortytwo
mon: set_healthy_stretch_mode in update_from_paxos, not random leader…
aea519d 
… calls!

Add header comment describing how this works now.

Fixes: https://tracker.ceph.com/issues/50308

Signed-off-by: Greg Farnum <gfarnum@redhat.com>
(cherry picked from commit adbaeee)

Resolves: rhbz#1949166
@gregsfortytwo
mon: MMonProbe: direct MMonJoin messages to the leader, instead of th…
3694ed8 
…e first mon

When monitors are joining a cluster, they may send an MMonJoin message to place
themselves correctly in the map in either handle_probe_reply() or
finish_election(). These messages must be sent to the leader -- monitors do not
forward each other's messages.

Unfortunately, this scenario was missed when converting the monitors to support
connectivity-based elections, and they're sending these messages to
quorum.begin(). Fix this by including an explicit leader in MMonProbe (that the
new monitor may reference in handle_probe_reply) and using the leader
value in both locations.

Fixes: https://tracker.ceph.com/issues/50345

Signed-off-by: Greg Farnum <gfarnum@redhat.com>
(cherry picked from commit ffa7ff3)

Conflicts:
	src/messages/MMonProbe.h

Signed-off-by:  Greg Farnum<gfarnum@redhat.com>

Resolves: rhbz#1939766
@idryomov idryomov changed the base branch from nautilus to nautilus-saved April 20, 2021 14:32
@idryomov idryomov changed the base branch from nautilus-saved to nautilus April 20, 2021 14:32
@jdurgin jdurgin changed the base branch from nautilus to nautilus-saved May 14, 2021 21:59
@jdurgin jdurgin changed the base branch from nautilus-saved to nautilus May 14, 2021 21:59
@github-actions
Copy link

github-actions bot commented Jun 3, 2021

This pull request can no longer be automatically merged: a rebase is needed and changes have to be manually resolved

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers
No reviews
Assignees
No one assigned
Labels
DNM needs-rebase
Projects
None yet
Milestone
nautilus
6 participants
@gregsfortytwo @neha-ojha @tchaikov @smithfarm @athanatos @ronen-fr