flaky: TestArchivalBlobSync in nodebuilder/tests

[rootulp]

Flaky test

Test: TestArchivalBlobSync
Package: github.com/celestiaorg/celestia-node/nodebuilder/tests
PR: #4956
CI run: https://github.com/celestiaorg/celestia-node/actions/runs/24693631027/job/72222790850

Error

prune_test.go:154:
    Error Trace:	/home/runner/work/celestia-node/celestia-node/nodebuilder/tests/prune_test.go:154
    Error:      	Should be false
    Test:       	TestArchivalBlobSync
...
network.go:78: tearing down testnode
network.go:119: error when closing API server close tcp 127.0.0.1:45623: use of closed network connection
FAIL	github.com/celestiaorg/celestia-node/nodebuilder/tests	30.103s

Notes

This test failure appears to be a flake — #4956 only changes GitHub Actions workflow YAML files and cannot affect Go integration tests. The assertion at prune_test.go:154 fails many times in the same run, and the test was previously observed failing on an unrelated PR (#4949). Retrying CI produced the same failure.

[rootulp]

Another recurrence today (2026-05-14):

• PR fix(api)!: set request-size limit and write/idle timeouts on JSON-RPC server #4997 — failed pruning run
• PR docs: add executable code walkthrough #4929 — failed pruning run

Both with the same error:

--- FAIL: TestArchivalBlobSync (30.03s)
    network.go:115: error when closing API server close tcp 127.0.0.1:XXXXX: use of closed network connection
FAIL    github.com/celestiaorg/celestia-node/nodebuilder/tests  30.0s

Neither PR touches nodebuilder/tests — confirmed flake. Re-ran failed jobs on both.

[neyy91]

Did some digging into the recurrences. The fix from #5017 holds on main — no
pruning-job failures there since it was merged on May 27. All failures since
then are on the v0.31.x release line, which doesn't have the backport: 4
failures on June 12 alone (e.g. the v0.31.3-mocha and v0.31.3-arabica release
runs), all with the pre-#5017 signature (prune_test.go:154: Should be false,
30s timeout).

The error when closing API server message in the logs is a red herring — it's
a t.Logf in testnode teardown (celestia-app) and never fails the test.

Opened a backport of #5017 to v0.31.x: #5053
This issue can probably be closed once the backport lands.