Message-ID: <gh-pgjdbc-pgjdbc-3895-c3953337277@github.com>
From: "vlsi (@vlsi)" <noreply+vlsi@github.com>
To: "pgjdbc/pgjdbc" <noreply+pgjdbc-pgjdbc@github.com>
Date: Tue, 24 Feb 2026 16:36:21 +0000
Subject: Re: [pgjdbc/pgjdbc] PR #3895: implement require_auth, this is pretty much how libpq does this. 
In-Reply-To: <gh-pgjdbc-pgjdbc-3895@github.com>
References: <gh-pgjdbc-pgjdbc-3895@github.com>
List-Id: <gh-pgjdbc-pgjdbc.github.com>
X-GitHub-Author-Login: vlsi
X-GitHub-Comment-Id: 3953337277
X-GitHub-Comment-Type: issue_comment
X-GitHub-Issue: 3895
X-GitHub-Repo: pgjdbc/pgjdbc
X-GitHub-Type: comment
X-GitHub-Url: https://github.com/pgjdbc/pgjdbc/pull/3895#issuecomment-3953337277
Content-Type: text/plain; charset=utf-8

Why 3fab588 is the first bad commit

The commit does two things:

1. Adds `autosave=always` to the CI test matrix
  Before this commit, tests never ran with autosave=always.
  The unlucky random combination assigned autosave=always to the PG 9.1 job.

2. Changes shouldCreateAutomaticSavepoint() to create more savepoints

The old code checked isTransactionActive() before checking AutoSave.NEVER, meaning it would bail out early in some cases. The new code reorders the checks and also expands isSpecialQuery() — but the net effect is that with autosave=always, every non-special statement now gets wrapped in SAVEPOINT / RELEASE  SAVEPOINT.

The combined effect

DatabaseEncodingTest.encoding() does ~11,100 INSERTs in one transaction. With autosave=always, each INSERT creates a subtransaction. PG 9.1's default max_locks_per_transaction=64 gives only ~6,400 lock table slots — far fewer than the ~11,100 needed.

---

Frankly, I think the test did the right thing: it identified a bug.

Then we could have several workarounds:
  1. Try if increase max_locks_per_transaction in the CI PostgreSQL 9.1 config helps (e.g., -c max_locks_per_transaction=256 in entrypoint.sh).
  2. Exclude autosave=always from PG 9.1 in the matrix — add a rule like:
  `matrix.imply({pg_version: '9.1'}, {autosave: {value: 'never'}});`
  3. Add periodic commits in `DatabaseEncodingTest` to keep subtransaction count manageable
  
I am leaning towards 1 or 2.