A throttled DynamoDB global secondary index throttles writes to the base table it indexes

With provisioned capacity, if a GSI has exhausted its write capacity the write to the base table is rejected too, because DynamoDB will not let an index fall arbitrarily behind. The symptom is ProvisionedThroughputExceededException on PutItem against a table whose own consumed capacity graph is nowhere near its limit.

Confirm it in CloudWatch by looking at WriteThrottleEvents with the GlobalSecondaryIndexName dimension rather than only the table dimension, which is where most dashboards stop.

The usual root cause is a low-cardinality index partition key, such as a status field with three values, which concentrates all index writes on a few partitions. The other is provisioning an index below the table on the assumption that fewer items project into it. Fixes are to over-provision the index relative to the table, move the table to on-demand which scales the index too, choose a higher-cardinality index key, or make the index sparse, since an item with no value for the index partition key is never written to it.