[
https://issues.apache.org/jira/browse/HDDS-17133?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18128904#comment-18128904
]
Siyao Meng commented on HDDS-17133:
-----------------------------------
Thanks [~echonesis].
1. One representative 5-node test per area is good enough. Running whole 3-node
suites with 5 too costs too much CI time. Use \{3, 5} only for single cheap
methods, as in {{TestSecretKeysApi}} (HDDS-17123).
2. It is a good idea to test with 2 voting nodes down with a 5 quorum. Can also
check that 3 down fails safely.
* Finalization: both
* Follower reads: both. for lease enabled, 2 down -> reads should work; 3 down
-> no stale reads. Then one lease-disabled case
* Listeners: 5 voters + 2 listeners, stop 3 voters -> no leader
* Snapshots: both. could be in one test
* Secure OM: tokens and S3 secrets stay valid across failovers
* Recon, S3 Gateway: one test each (lower priority)
3. +1 to separate PRs. Pls feel free to one sub-task per area under HDDS-17122.
4. One concern is CI time addition. It might not be feasible with the GH CI
resources given to run all those suites every time. So we might have to leave
some parts disabled for regular post-commits, and maybe run them once in a
while so they don't rot (or maybe even consider replacing some 3 nodes test
with 5 nodes, provided that it doesn't add too much CI time. That is up for
discussion).
> Extend 5-node HA test coverage to upgrade finalization, Recon, S3 Gateway,
> follower read, listener OMs and secure OM
> --------------------------------------------------------------------------------------------------------------------
>
> Key: HDDS-17133
> URL: https://issues.apache.org/jira/browse/HDDS-17133
> Project: Apache Ozone
> Issue Type: Sub-task
> Components: OM HA, SCM HA, test
> Reporter: Siyao Meng
> Priority: Major
>
> The new 5-node integration tests cover the Ratis basics. Not yet covered with
> more than 3 voting nodes:
> * OM and SCM upgrade finalization (TestOMUpgradeFinalization,
> TestScmHAFinalization use 3).
> * Recon syncing from a 5-OM ring and following OM leader changes.
> * S3 Gateway failover across 5 OMs.
> * OM follower read (LINEARIZABLE / leader lease) with 5 OMs.
> * Listener OMs together with 5 voting OMs.
> * FSO/OBS filesystem snapshots and their install on lagging OMs in a 5-OM
> ring.
> * Secure (Kerberos) 5-OM cluster.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
