Skip to content

Not for merge: CI proof (log-target mutation) - #4505

Closed
erikdarlingdata wants to merge 8 commits into
devfrom
ci-proof/log-target-mutation
Closed

erikdarlingdata wants to merge 8 commits into
devfrom
ci-proof/log-target-mutation

Conversation

@erikdarlingdata

Copy link
Copy Markdown
Owner

Not for merge. A CI proof for #4504: one commit on top of it drops LOG from PgLogEntryAssembler's label alternation. Expected: the two self-hosted log end-to-end tests (and the parser pins) go RED in the Darling PostgreSQL shards. Closed and its branch deleted once CI reports.

…requests

Adds a fifth throwaway log-format cluster (port 5545, logging_collector on,
log_connections/log_disconnections/log_lock_waits on, a %u@%d-bearing
prefix, UTC timezone) to build.yml's darling-pg job, plus a deterministic
workload step that produces one real error, connection, and lock-wait
event before the tests run. Sets DARLING_TEST_PG_LOG_TARGET and
DARLING_TEST_PG_UTC_LOG_TARGET on the test step so both previously-skipped
live end-to-end tests in PgLogEventsPipelineTests run on every pull
request. nightly.yml starts no log-format clusters today, so it is left
unchanged.
Not for merge. Proves the CI wiring on the sibling branch is real: with
this line dropped, PgLogEntryAssembler.Assemble stops recognising every
LOG-severity server line, which is most of what the connection and
lock-wait families read (both parsers require Severity == LOG). Both new
end-to-end tests in PgLogEventsPipelineTests must go red under this
change, because the log lines their workload produced can no longer be
folded into entries at all.
psql only accepts libpq keywords or -h/-p/-U/-d flags, not an
Npgsql-style connection string; the step's calls used the latter
and failed with invalid connection option 'Host'. The tests'
own env vars (Npgsql form, read by Npgsql) are unchanged.
Outside a transaction block, PostgreSQL rejects LOCK TABLE immediately, so the waiter session never actually waits for the lock. Wrap it in BEGIN/COMMIT to match the holder session. Also add -v ON_ERROR_STOP=1 to the psql calls that are meant to succeed, so a SQL error in them is no longer silently swallowed.
@erikdarlingdata

Copy link
Copy Markdown
Owner Author

Closing: CI proof recorded on #4504.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant