docs truth: a shipped version claimed a change it never made

[jacobecox]

Three documentation corrections, all in place — helm template output is byte-identical for each, verified.

Found by the docs writers treating the chart as source of truth rather than the README, which is exactly what that rule exists for.

1. postgres-multi-location 1.1.0's headline change never shipped

Its README says "1.1.0 widens the tolerance" and describes moving worst-case failover from 45 to 60 seconds. The chart ships ttl: 45 / loop_wait: 10 / retry_timeout: 15 — and so do 1.0.3 and 2.0.0, byte for byte:

1.0.3    ttl:45 loop_wait:10 retry_timeout:15
1.1.0    ttl:45 loop_wait:10 retry_timeout:15
2.0.0    ttl:45 loop_wait:10 retry_timeout:15

The README's own edit-config remediation command sets 45/10/15, contradicting the paragraph three lines above it, and the published CHANGELOG entry repeats the claim.

So a user reading either doc believes their cluster tolerates a 60-second etcd outage when it gives up at 15. That is a false operational guarantee about failover, not a doc nit.

The README now states what is actually shipped and frames widening as a deliberate trade left to the operator, with Patroni's loop_wait + 2 * retry_timeout <= ttl constraint. The CHANGELOG entry carries a dated correction rather than being silently rewritten — it was published and people may have read it.

2. grafana-multi-location 2.0.0 tells users about a key that no longer exists

values.yaml still said a location "must be one of global.gvc.locations" — a key the conversion removed. Comment-only.

3. cpln-advisor 2.0.0's restore caveat is now false, and the command is unsafe

The README says the restore "has not been executed against this template". The test round executed it end to end — 90 KB dump, 8 tables, database dropped, restored, application working on the restored data — so the docs page and the README now disagree, with the README being the stale one.

It also needs a requirement the round found the hard way: the dump comes from postgres:18, whose pg_dumpall emits \restrict directives that older clients reject while psql exits 0. It succeeded only because those lines were not load-bearing. The command now carries -v ON_ERROR_STOP=1, because a silently-successful restore is the dangerous case.

🤖 Generated with Claude Code