Tune recomputeThumbnail bulk action to avoid transaction timeout or stream failure

Today the recomputeThumnail action is configured like this

   <action name="recomputeThumbnails" inputStream="bulk/recomputeThumbnails" bucketSize="100" batchSize="10"
      httpEnabled="false" />
...
      <policy name="default" maxRetries="3" delay="1s" maxDelay="10s" continueOnFailure="true" />

this means 100 documents per records with batch of 10 docs per transaction
The transaction nuxeo.db.transactiontimeout is 5min by default, if thumbnail conversions cannot be done within 5min there is a transaction timeout that triggers a retry so the record is reprocessed, potentially this will generate again a timeout while processing the same batch of documents.

At best the record is skipped and we miss some conversions, at worst we failed to respect the kafka poll interval and this creates a stream failure which requires a restart of the node.

possible solution

• assume that the processing to generate the thumbnail is slow and reduce the batch size to 1
• in the case of a transaction timeout make sure that the bulk command status is completed with an error status.