search

IBM / kar

KAR: A Runtime for the Hybrid Cloud
Apache License 2.0
28 stars 11 forks source link

bug fix in actor scheduling for recovery from a blocked self-call #318

Closed dgrove-oss closed 2 years ago

dgrove-oss commented 2 years ago

During recovery, when an message is identified as needing to be blocked to wait for a child call to complete it is necessary to eagerly set the activeFlow on the freshly created actor instance. This recreates the locking state before recovery and allows re-entrant calls to be properly identified as they are dequeued from kafka.

Also extend fault-tolerance-actor test case to include self-calls.