Description:
his is general question / feature request
Cluster allows a number of resource limits to be configured via the config.ini file, but provides very little visibility of resource utilisation at runtime. This makes it hard to validate the parameters chosen, as there is no feedback until there is a failure due to resource exhaustion.
Some resources (i.e. memory) can be monitored through the use of the ndb_mgm DUMP command, but there is little information on what can be monitored, and what the numbers mean.
The sort of resources that we would like information on include :
- REDO buffer usage (current, high water mark and average)
The in-memory buffer which buffers REDO logs en-route to disk
- REDO fragment log file usage (current, high water mark and average)
The circular buffer on disk which all REDO must fit into
- Connection objects, Transaction objects, Operations, Scan operations etc. (Current and HWM)
- Number inactive transaction rollbacks
- Number 'deadlock' / lock claim timeout transaction rollbacks
- HWM of concurrent index scans
(We do a lot of index scans and are concerned that we may approach the system limit here)
etc...
If it possible to monitor any of these resources via ALL DUMP commands then we would be interested in the details...
Thanks
How to repeat:
.
Suggested fix:
.
