test: fix read-replica and mongo seed flakes in integration tests - #17543
Draft
denolfe wants to merge 5 commits into
Draft
test: fix read-replica and mongo seed flakes in integration tests#17543denolfe wants to merge 5 commits into
denolfe wants to merge 5 commits into
Conversation
…ica flake payload.db.drizzle is the withReplicas proxy, so .select() is routed to a read replica. The vector tests write rows through payload.create and then read them back with the raw drizzle handle, which misses rows that have not replicated yet and fails with an empty result set. Payload's own reads avoid this via shouldReadFromPrimary, which keeps reads on the primary for readReplicasAfterWriteInterval after a write. The raw handle bypasses that guard, so query primaryDrizzle directly instead.
The cart tests read productId/variantId with optional chaining and cast the result, so an incomplete seed leaves them undefined without failing the hook. The guest cart secret has the same problem: when it is absent every later request fails access control and reports "cart not found". Both cases currently surface as thirteen unrelated-looking failures well after the real problem. Assert on them where they are read instead.
…o flake The seed created all variant options for a variant type with Promise.all. On the first write the variantOptions namespace and its versions namespace do not exist yet, and MongoDB cannot create the same namespace from two concurrent transactions, so the parallel creates raced and one commit failed with a TransientTransactionError WriteConflict. The seed then swallowed that error and returned false, which onInit ignores, so Payload booted with empty collections and thirteen cart tests failed much later with unrelated messages. Create the options sequentially and rethrow instead.
Contributor
📦 esbuild Bundle Analysis for payloadThis analysis was generated by esbuild-bundle-analyzer. 🤖 |
Mongoose starts building indexes as soon as the connection opens and does not wait for those builds to finish. They ran at the same time as the test-only database drop and the first writes from onInit, so MongoDB rejected the writes with "Cannot create collection - database is in the process of being dropped" and "Unable to write to collection due to catalog changes". The documentdb int shard failed this way in collections-graphql. Disable autoIndex when the database will be dropped, then build the indexes after the drop and wait for them. The awaited build now also covers version models, which were previously left to the background builds and so had no indexes when connect resolved. The shared globals discriminator model stays excluded because its auto-named indexes collide with each other.
…cart tests The authenticated user cart and cart transfer beforeAll hooks still used unguarded optional chaining for productId, and several guest cart creations never asserted the secret came back.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Fixes two integration test flakes that fail on
mainand on feature branches. Both defects are in the test code, not in Payload.Root Cause
The postgres vector tests read through
payload.db.drizzle. That handle is thewithReplicasproxy (drizzle's replica router), so a raw.select()goes to a read replica. The tests write rows withpayload.createand read them back immediately. Rows that have not replicated yet are absent. The query returns nothing. Payload's own reads stay correct throughshouldReadFromPrimary, which pins reads to the primary forreadReplicasAfterWriteIntervalafter a write. The raw handle skips that guard.The ecommerce seed created each variant type's options with
Promise.all. On the first write, thevariantOptionsnamespace (MongoDB's term for a collection) and its versions namespace do not exist yet. MongoDB cannot create the same namespace from two concurrent transactions, so one commit fails with aTransientTransactionErrorwrite conflict. The seed caught that error and returnedfalse.onInitignores the return value, so Payload booted with empty collections. Thirteen cart tests then failed with messages that point nowhere near the seed.Key Changes
Postgres vector tests query the primary
primaryDrizzlewhen it is set. They fall back todrizzlewhen no replica is configured.Ecommerce seed creates variant options sequentially
payload.logger.errorand rethrows. A broken seed fails at the seed rather than during the cart tests.Ecommerce cart fixtures fail fast
beforeAllhooks throw when the seed yields no product or variant. Optional chaining previously left those ids undefined and the hook still passed.