🐛(backend) do not fail the parent path index migration on an existing index

[NathanVss]

Purpose

Until item_parent_path_idx is valid, the queries relying on it read the whole item table. On preprod the new code served requests while the index was still being built: item creation took 17s and the migration job died after 33 minutes, leaving an invalid index.

On large databases the index must therefore be created by hand before the release is deployed, but the migration then fails with relation "item_parent_path_idx" already exists.

Proposal

Migration 0031 now uses a reusable AddIndexConcurrentlyIfMissing operation (core/utils/migrations.py), which looks the index up before building it:

• no index: build it, as before
• valid index: nothing to do
• invalid index (interrupted concurrent build): drop it and build it again

Deploying on a large database

Before deploying, run and check that indisvalid is true:

CREATE INDEX CONCURRENTLY "item_parent_path_idx"
ON "drive_item" ((subpath("path", 0, nlevel("path") - 1)));

SELECT indisvalid FROM pg_index
WHERE indexrelid = 'item_parent_path_idx'::regclass;

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud