Unsolved
This post is more than 5 years old
5 Posts
0
3628
nsrjobd - indexes
We have an occassional problem where it appears Networker gets overloaded - specifically nsrjobd.
By looking at the daemon log you can see that everything seems to have ground to a halt. Hardly any messages are appearing.
We then get several messages like - 39078 12/01/11 01:03:36 nsrjobd SSNCHNL error: Aborting session channel connection (e6178) to 10.2.25.124; why = There is no process to read data written to a pipe.
Then one of two things happens , either
1) Several groups are then aborted with the error message "index was never started". After this we normally have to restart Networker as it doesn't seem to recover.
or
2) Networker writes out lots of indexes and completes the savegrps successfully and returns to normal running.
I'm really interested to understand why Networker is apparantly holding off from writing indexes when the savegrps seem to have completed and then suddenly writes loads at once. I'm guessing that there's some condition or resource that it's waiting for ?
Also, if the are any commands , either Networker ones or even OS (AIX Unix) that I could use to check if nsrjobd or any other process is getting over worked.
Thanks
squesnelle
15 Posts
0
January 12th, 2011 10:00
The indexes are written as the backups are done, but when a client is complete, savegroup will usually do a backup of the indexes for the client.
This save of the indexes may be where you are encountering the problem. Especially if the server is heavily loaded. I would look at the value of the client parallelism for the server machine, this controls how many concurrent saves will be on that host at any given time. Also if you aren't in the 7.5.4 or 7.6.1 line then I would recommend upgrading to those recent builds.
Alternatively, you can disable saving the indexes as part of your normal savegroups, and instead choose a quiet time during your day to do the basic maintenance of running nsrim -X, then backup the bootstrap, indexes for all clients at that time.