MySQL Bugs: #16145: MySQL Cluster Clusterlog definitions

Bug #16145	MySQL Cluster Clusterlog definitions
Submitted:	3 Jan 2006 3:53	Modified:	7 Apr 2006 10:12
Reporter:	Adam Dixon (Candidate Quality Contributor)	Email Updates:
Status:	Closed	Impact on me:	None
Category:	MySQL Server: Documentation	Severity:	S3 (Non-critical)
Version:	5.0	OS:
Assigned to:	Jon Stephens	CPU Architecture:	Any

Description:
Place documentation of Clusterlog output into docs;

eg (From Mikael Ronström, cluster list)

TransCount = Number of transactions with this node as coordinator
CommitCount = Number of transactions committed with this node as
coordinator
ReadCount = Number of primary key reads
SimpleReadCount = Number of primary key reads reading the latest
committed value
WriteCount = Number of primary key writes (insert, update, delete)
AttrInfoCount = Number of words to describe reads and writes received
ConcurrentOperations = Number of concurrent operations ongoing at the
moment of report
AbortCount = Number transactions aborted with this node as coordinator
Scans = Total number of scans
Range scans = Index scans

> 2005-12-23 12:26:34 [MgmSrvr] INFO     -- Node 12: Operations=20
> ...
> 2005-12-23 12:26:34 [MgmSrvr] INFO     -- Node 14: Trans. Count = 11,
> Commit
> Count = 10, Read Count = 9, Simple Read Count = 0,
> Write Count = 10, AttrInfo Count = 498, Concurrent Operations = 0,
> Abort
> Count = 1
>  Scans: 6 Range scans: 0
> ....

These are statistics generated by the transaction coordinator module.

> 2005-12-23 12:26:36 [MgmSrvr] INFO     -- Mean loop Counter in doJob
> last
> 8192 times = 17

The ndbd has a scheduler that runs in an eternal loop. In each loop it
1) Reads incoming messages from sockets into job buffer
2) Check if any timed messages are to be executed, if so put them in
job buffer
3) Execute in a loop messages in the job buffer
4) Send any distributed messages generated by executing the messages in
the job buffer
5) Wait for new incoming messages

This statistic report the number of loops in 3). The reason to check
this is that if one can make this bigger
then the buffer on the TCP/IP part gets to be better. So one can check
what goes on when adding threads
and MySQL Servers to the cluster, hopefully this figure should go up
then.

> ....
> 2005-12-23 12:37:33 [MgmSrvr] INFO     -- Node 13: Mean send size to
> Node =
> 2 last 4096 sends = 6007 bytes
>

This reports the mean size of writes on the TCP socket for
communicating between
Node 13 and Node 2 in this case (higher value means less cost per byte
sent,
maximum is around 64 kB).

There is also a similar one for receive size.

How to repeat:
ndbmgm> all clusterlog statistics=15

Suggested fix:
Place info in documentation.

Thank you for a documentation reqiest. This page, http://dev.mysql.com/doc/refman/5.0/en/mysql-cluster-logging-management-commands.html, should be extended or, maybe, a new page added to the manual.

Thank you for your bug report. This issue has been addressed in the
documentation. The updated documentation will appear on our website
shortly, and will be included in the next release of the relevant
product(s).

Additional info:

Added new Cluster Log Statistics subsection to Cluster Management section, will be available as http://dev.mysql.com/doc/refman/4.1/en/mysql-cluster-log-statistics.html /
http://dev.mysql.com/doc/refman/5.0/en/mysql-cluster-log-statistics.html /
http://dev.mysql.com/doc/refman/5.1/en/mysql-cluster-log-statistics.html