Details
-
New Feature
-
Status: Closed (View Workflow)
-
Minor
-
Resolution: Fixed
-
1.1.3
-
None
-
Multi server separated out in two data centers and geographic locations
Description
According to the InfiniDB Concepts Guide, User and Performance Modules can be separated out in different data centers and geographic locations. Thus, I am hopeful that MariaDB ColumnStore could provide such feature.
As linuxjedi informed that there were several bugs in the network packet handling and compression that are fixed in the develop-1.1 branch (which will make up the 1.1.3 release), I deployed MariaDB ColumnStore from 1.1.3 repository (build of Jan 2) on three combined UM/PM nodes, with PM3 in a different geo-location separated by 25 ms of RTT from PM1/2 nodes. The system is loaded with 3M rows of data separated across all three PMs.
After a system start, a test data aggregation query works as expected and completes in 15-20s. However, after few days of system idling and no records in the logs, the same data aggregation query fails with
ERROR 1815 (HY000): Internal error: DistributedEngineComm::write: Broken Pipe error
|
having records in the /var/log/mariadb/columnstore/crit.log
Jan 8 13:31:42 pm-01 joblist[17132]: 42.993173 |24|0|0| C 05 CAL0000: DistributedEngineComm::write: Broken Pipe error
|
Jan 8 13:31:42 pm-01 joblist[17132]: 42.993849 |0|0|0| C 05 CAL0000: /data/buildbot/bb-worker/deb-ubuntu16/build/mariadb-columnstore-server/mariadb-columnstore-engine/dbcon/joblist/distributedenginecomm.cpp @ 382 DEC: lost connection to xx.xx.xx.xx
|
The same query executes fine following system shutdown/start.
How can I increase the log level verbosity to debug broken pipe and lost connection issues?