MongoDB | 2592246

Info

This error surfaced while testing FLE state collections residing on arbitrary data shards but it looks like a generic problem: batched command responses containing write errors are never serializing error labels (despite being able to deserialize them). Scenario: Create FLE collection on primary shard and FLE state collections on a different shard (this will cause FLE cruds to be executed in distributed transactions targeting 2 different shards) Issue a write within a transaction on the encrypted collection A StaleConfig error is returned. Despite being a transient transaction error, the client will not retry the transaction because the TransientTransactionError label is not attached to the reply. Attaching a reproducible to apply on the top of commit 2f3dd3bb7d and execute any test with FLE transactions to verify that the reply is missing "errorLabels":["TransientTransactionError"]. For example, executing this: ./buildscripts/resmoke.py run --suite fle2_sharding --storageEngine=wiredTiger --jobs=1 --storageEngineCacheSizeGB=0.5 --dbpath=/tmp/testpath --runAllFeatureFlagTests src/mongo/db/modules/enterprise/jstests/fle2/txn_insert.js This is the returned error: { "nInserted" : 0, "writeError" : { "code" : 13388, "errmsg" : "Transaction 92f84da6-dbee-42b5-8df7-362955a88b51 - 47DEQpj8HBSa+/TImW+5JCeuQeRkm5NMpJWZG3hSuFU= - - :0 was aborted on statement 2 due to: an error from cluster data placement change :: caused by :: Encountered error from localhost:20001 during a transaction :: caused by :: sharding status of collection txn_insert.enxcol_.basic.ecoc is not currently known and needs to be recovered" } }

Top User Comments

Steps to Reproduce