Alexey Goncharuk created IGNITE-7692:
Summary: affinityCall and affinityRun may execute code on backup
Issue Type: Bug
Affects Versions: 2.1
Reporter: Alexey Goncharuk
Fix For: 2.5
Apparently, the affinityCall and affinityRun methods reserve partitions and
check their state to be OWNING, however, if topology changes and partition role
is changed to backup from primary, the code is still executed.
This can be an issue if a user executes a local SQL query inside the
affinityCall runnable. In this case, the query result may return null.
This can be observed in the
IgniteCacheLockPartitionOnAffinityRunTest#getPersonsCountSingleCache - note an
additional check I've added to make the test pass.
I think it is ok to have an old semantics for the API, because in some cases
(scan query, local gets) a backup OWNER is enough. However, it looks like we
need to add another API method to enforce that affinity run be executed on
primary nodes and forbid primary role change.
Another option is to detect a topology version of the affinity run and use that
version for local SQL queries.
This message was sent by Atlassian JIRA