Increasing disk redundancy chokes cored protocol
Details
| Detail name | Value |
|---|---|
| Changelog Number | 28060 |
| Type | Bug |
| Status | Resolved |
| Affected Versions | Exasol 8.0.0 |
| Fix Versions | Exasol 2025.1.9, Exasol 2026.1.0, Exasol 2025.2.1 |
| Resolution Date | 2026-02-18 |
Description
Increasing a volume’s redundancy will make the system unresponsive. This is caused by a background recovery optimization that broadcasts many large (~160KiB) events using the cored protocol. The number of messages generated greatly overcomes the throughput, and collapses the protocol queues. Prior to CHANGELOG: During extremely high load the cluster may become unstable, this would manifest in intermittent database crashes and cluster instability (coming in and out of configurations). With CHANGELOG: During extremely high load the cluster may become unstable in place, the cluster remains stable, but applications (like csinfo) will fail to run, as they send events that are queued and “never“ processed (the queue is so large that it would take hours to empty).
Workaround
There is no workaround.
Fix
These large messages are not sent. They are not required by the storage service for normal function. The system will remain stable even under very heavy load and an increase of redundancy happening at the same time.
