Public Cloud: Short network splits misinterpreted as node crashes
Details
| Detail name | Value |
|---|---|
| Changelog Number | 9932 |
| Type | Bug |
| Status | Resolved |
| Affected Versions | Exasol 6.1.0, Exasol 6.2.0, Exasol 7.0.0-alpha1 |
| Fix Versions | Exasol 6.2.7, Exasol 7.0.alpha2, Exasol 6.1.10 |
| Resolution Date | 2020-04-16 |
Problem
When running affected Exasol versions on public cloud hardware, certain events in the cloud infrastructure can be misinterpreted as node failures.
These events include but are not limited to:
- Updates to the hypervisor that preserve the running virtual machine.
- Changes to the cloud provider's network infrastructure.
In these circumstances one or multiple nodes can be unreachable for the other nodes in the cluster long enough to cause Exasol's failover handling to trigger and consider the nodes crashed.
However, because the unreachable nodes will be reachable from the cluster again after the interruption is over a restart of the database is unnecessary if this condition can be detected correctly.
Workaround
Exasol Support can supply affected customers with a hotfix patch that mitigates the issue.
FIx
For future updates of supported Exasol releases, we are going to improve internal timeout logic to make this misidentification of crashed nodes less likely to occur.
