Install pgBackRest and add initial backup and rewind settings #32
Conversation
Codecov Report
@@ Coverage Diff @@
## main #32 +/- ##
==========================================
+ Coverage 65.35% 66.19% +0.83%
==========================================
Files 6 6
Lines 814 840 +26
Branches 115 121 +6
==========================================
+ Hits 532 556 +24
- Misses 258 260 +2
Partials 24 24
Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here. |
marceloneppel
changed the title
marceloneppel
changed the title
src/cluster.py
Outdated
| for attempt in Retrying(stop=stop_after_delay(60), wait=wait_fixed(3)): | ||
| with attempt: | ||
| cluster_status = requests.get( | ||
| f"{self._patroni_url}/cluster", |
There was a problem hiding this comment.
A lot of these URL magic strings are similar - can you extract them into variables please?
There was a problem hiding this comment.
Changed to a constant on 5adae46.
tests/integration/helpers.py
Outdated
| apps=[application_name], | ||
| status="active", | ||
| timeout=1000, | ||
| timeout=2000, |
There was a problem hiding this comment.
This isn't really relevant to this ticket, but should adding a relation really take half an hour? It might be worth looking into why this takes so much time
There was a problem hiding this comment.
Good catch! I think this change was made just for a local test and I forgot to remove it. Reverted on 0bb1414.
it should be switched to pgBackRest only after initialization of the pgBackRest stanza (in other words after storage initialization). |
…cal#32) * Add pgbackrest * Add initial backup settings * Add additional backup settings * Remove settings * Rename user * Add test for TLS being used on pg_rewind connections * Remove table creation * Readd write to the database * Change bootstrap contraints * Change the way the service is stopped * Add one more call to service stop * Change the way the service is stopped * Change systemd unit * Increase test timeout * Remove test code * Read code * Readd code * Change WAL trigger mechanism * Fix check * Improve code * Remove instance promotion * Readd instance promotion * Change test retry logic * Remove debug calls * Add replica reinitialization * Change checks order * Add reinitialize call * Improve reinitialize call * Remove unused code * Pin OS on release workflow * Change whitelist_externals to allowlist_externals * Change whitelist_externals to allowlist_externals * Add API request timeout * Add unit tests * Remove log * Revert timeout * Extract endpoint from URL to constant
5 participants
Issue
Solution
Install pgBackRest.
Add the following settings for backup operations:
achieve_mode=on
archive_command=/bin/true
wal_level = logical
Add the following settings for rewind operations:
remove_data_directory_on_rewind_failure = true
remove_data_directory_on_diverged_timelines = true
Add a user for pg_rewind and test that it uses TLS when connecting to other PostgreSQL instance.
Context
This PR is almost a copy of Add initial backup and rewind settings postgresql-k8s-operator#45.
The main differences are:
A call to reinitialise the replica was added to
src/charm.py(because at some points in some tests the replica became too far behind the timeline of the primary and this is the recommended way to solve that is to reinitialise the replica data directory).On
src/cluster.pya call to the Patroni API to retrieve the lag information. Also, timeouts were added to the API calls to avoid long requests that make the cluster slow, specially in the freeze db process test (that was failing sometimes).Testing
tests/unit/test_patroni.pywere updated to match the new settings.pg_rewindis using TLS.Release Notes