[laxman-ch]

Problem

GitHub is migrating the ubuntu-latest label from 24.04 to 26.04. The rollout is staged, so jobs currently land on either image depending on which runner they get — CI is non-deterministic rather than simply broken. (The runner-images README still documents ubuntu-latest as 24.04, while our runs are already getting 26.04.)

Nothing in the repo changed. The same job passed on an identical tree three days before it started failing — only the image differs:

Date	Run	Image	full-build-jdk11
2026-10-16	38856125551	ubuntu-24.04	pass
2026-10-19	39272997735	ubuntu-24.04	pass
2026-10-22	39714174594	ubuntu-26.04	fail

This blocks every PR against branch-3.6 — both required checks are red, currently including #159.

Why pin rather than fix forward

26.04 breaks the C client in at least two unrelated ways.

1. Compile. strchr and friends are const-generic there, and the build uses -Werror:

src/cli.c:503:13:      error: assignment discards 'const' qualifier from pointer target type
src/cli.c:709:11:      error: assignment discards 'const' qualifier from pointer target type
src/zookeeper.c:1263:18: error: assignment discards 'const' qualifier from pointer target type

2. SSL tests. With the compile errors patched locally, the build gets further and then fails in cppunit — 2 of 83 tests, ZSSLCONNECTIONERROR, against 26.04's newer OpenSSL versus what ssl/gencerts.sh produces:

TestClient.cc:813         Expected: 0, Actual: -15   (Zookeeper_simpleSystem::testSSL)
TestReadOnlyClient.cc:118 Expected: 0, Actual: -15

Those are two different root causes — a compiler change and a crypto-tooling change — and there is no basis for assuming the second is the last. Certifying the C client and its test suite on 26.04 is its own piece of work, and doing it a failure at a time through CI round-trips is the wrong place for it.

So this PR restores a reproducible green baseline and leaves 26.04 as a deliberate, separate effort.

Change

ubuntu-latest → ubuntu-24.04 in the three active workflows. ubuntu-24.04 is an explicitly supported label, not a deprecated one.

Workflow	Jobs	Why
ci.yaml	all four full-build-*, incl. both required checks	the actual breakage
e2e.yaml	compatibility	currently green, pinned so it cannot drift next
ci-jfrog-workflow.yml	release deploy	release path; a surprise image change here is worse than in CI

ci-bintray-workflow.yml_old is left alone — the extension means Actions never loads it.

Follow-up

The 26.04 work is not lost. The compile fixes are written and were verified green on 26.04 in this PR's earlier revisions (dc9c6c9e3, f36a0e084) — cli.c drops a const that was never accurate (both sites write *ptr = '\0'), and zookeeper.c makes index_chroot properly const. They are a good starting point for the certification effort, which still needs the gencerts.sh/OpenSSL side.

Testing

Both required checks, plus e2e, should return to the 24.04 behaviour that was green through 2026-10-19.

[laxman-ch]

Changed approach — this PR is now the runner pin, not the C fixes.

The earlier revisions fixed the 26.04 compile errors directly, and that worked: both required checks went green. But the build then got further and failed in cppunit instead — 2 of 83 tests, ZSSLCONNECTIONERROR, from 26.04's newer OpenSSL versus ssl/gencerts.sh. A second, unrelated root cause, with no reason to think it was the last.

Certifying the C client on 26.04 is a separate effort rather than something to converge on one CI round-trip at a time, so this PR now just pins to ubuntu-24.04 to restore a reproducible green baseline. The compile fixes are kept at f36a0e084 and referenced in the description for that follow-up.