add linearizability nightlies for release 3.4/3.5 #14915
Conversation
tjungblu
force-pushed
the
linnightbranch
branch
2 times, most recently
from
bf5e312
to
ce33a8c
Compare
|
also looking into today's run: https://github.com/etcd-io/etcd/actions/runs/3646971615/jobs/6158650127 There's a 30m timeout on the test runner itself, we need to make this configurable too. EDIT: done. |
tjungblu
force-pushed
the
linnightbranch
branch
3 times, most recently
from
b3a16b4
to
8c7e5ac
Compare
This CL refactors the tests to reuse a single workflow that has parameters. This is then reused for PRs/pushes and the nightlies. Signed-off-by: Thomas Jungblut <tjungblu@redhat.com>
tjungblu
force-pushed
the
linnightbranch
branch
from
8c7e5ac
to
c321a59
Compare
| timeout-minutes: 180 | ||
| uses: ./.github/workflows/linearizability-workflow.yaml | ||
| with: | ||
| ref: release-3.4 |
There was a problem hiding this comment.
Older branches don't have code for linearizability tests. You need to compile old etcd version, but run linearizability tests latest main branch. Please see https://github.com/etcd-io/etcd/pull/14912/files
| ref: main | ||
| count: 500 | ||
| timeoutDuration: 3h | ||
| test-3.5: |
There was a problem hiding this comment.
Note, older releases don't have same gofailpoints. Tests will fail when trying to trigger a failpoint that is present on main which is not available on release-3.4 nor release-3.5.
We might need to skip some of scenarios that are not yet backward compatible.
To make new failpoints work with older releases we would need:
- merge remove path from failpoint gofail#44 (comment) that will ensure that failpoints have same name
- For linearizability tests implement checking which failpoints are available in etcd server (should be available in one of HTTP endpoints) and triggering only those available.
There was a problem hiding this comment.
Second point was implemented in #15046
| with: | ||
| go-version: ${{ inputs.goVersion }} | ||
| - run: | | ||
| make gofail-enable |
There was a problem hiding this comment.
Scripts for building on older branches are really different.
|
All dependent features needed for this feature were implemented (consistent naming of failpoints between releases, triggering only available failpoints). @tjungblu do you want to finish the PR? |
|
Superseded by #15106 |
This CL refactors the tests to reuse a single workflow that has parameters. This is then reused for PRs/pushes and the nightlies.
Signed-off-by: Thomas Jungblut tjungblu@redhat.com