"Couldn't write state file" problem...

Elphidieus
Elphidieus
Joined: 20 Feb 05
Posts: 245
Credit: 20,603,702
RAC: 0
Topic 193331

Hi...

Anyone of you Mac users out there having these error messages?

Mon Nov 19 22:21:13 2007||Can't rename client_state_next.xml to client_state.xml; check file and directory permissions
Mon Nov 19 22:21:13 2007||rename client_state_next.xml to client_state.xml returned error 2: No such file or directory
Mon Nov 19 22:21:13 2007||[error] Couldn't write state file: system rename

Currently on BOINC 5.10.29, crunching on Einstein Mac Intel 4.10 app. And my crunch time steadily increases for no apparent reason, from 18,000 sec to 24,000 sec.

Would appreciate if anyone can shed some light on this...

Thanks.

Jord
Joined: 26 Jan 05
Posts: 2,952
Credit: 5,878,802
RAC: 6,460

"Couldn't write state file" problem...

How are your permissions on the BOINC folder set? Does the account you run BOINC with have permission to write to disk?

Elphidieus
Elphidieus
Joined: 20 Feb 05
Posts: 245
Credit: 20,603,702
RAC: 0

Read and write via

Read and write via boinc_master of course, except others...

Jord
Joined: 26 Jan 05
Posts: 2,952
Credit: 5,878,802
RAC: 6,460

OK, check them

Message 75487 in response to message 75486

OK, check them anyway.

Here's what BOINC does several times per second: write its state to client_state_next.xml, then once every 10 to 60 seconds (and more if you have your write to disk amount set higher than default) it writes this changed state to client_state.xml

So you do not see a file called client_state_next.xml in your BOINC directory, but as long as BOINC runs it is there. When BOINC stops the final state from that file is written to client_state.xml (which in itself is backed up to client_state_prev.xml). All these files need full permission to write to disk, so the BOINC folder (and all sub-folders) need full write permission for whichever account you run BOINC with. Perhaps that something changed on the boinc_master or boinc_project accounts.

If permissions are without a doubt correctly set, try to stop boinc and restart it. That may get things going again.

Elphidieus
Elphidieus
Joined: 20 Feb 05
Posts: 245
Credit: 20,603,702
RAC: 0

Changing permission will

Changing permission will result in BOINC Manager itself refusing to start, prompting re-installation.

Stopping and restarting BOINC won't help either.

I've had less of these messages while I was with SETI compared to Einstein, so I'm not sure if it's project-related.

Bikeman (Heinz-Bernd Eggenstein)
Bikeman (Heinz-...
Moderator
Joined: 28 Aug 06
Posts: 3,522
Credit: 694,398,560
RAC: 136,298

RE: Changing permission

Message 75489 in response to message 75488

Quote:

Changing permission will result in BOINC Manager itself refusing to start, prompting re-installation.

Stopping and restarting BOINC won't help either.

I've had less of these messages while I was with SETI compared to Einstein, so I'm not sure if it's project-related.

I don't think this is project related. A re-install might help, it won't damage/erase any files so it's safe.

The gradual increase in runtime you described in your initial question is normal for S5R3: some units take longer than others, you will see crunching times fluctuating quite significantly. But on average the 4.10 beta app for Mac OS is still the fastest around.

CU
Bikeman

peanut
peanut
Joined: 4 May 07
Posts: 162
Credit: 9,644,812
RAC: 0

Maybe I can help ease your

Maybe I can help ease your mind about this issue.

From my experience, I would not worry about this "can't rename client_state_next.." thing. It does not seem to do any harm on my Mac Pro 8 core (looks like you have same thing Elphidieus).

I get many of those messages if all my cores seem to be "in sync" like when I first started my machine up and if a few WUs start at around the same time. After things get out of sync I get far fewer can't rename messages. You may have already noticed a decrease if over time your Tasks have drifted into various states of progress. My theory is that it might be a race like condition where two or more cores somehow try to access the client_state file.

Jord
Joined: 26 Jan 05
Posts: 2,952
Credit: 5,878,802
RAC: 6,460

I've asked Charlie Fenton,

I've asked Charlie Fenton, the Mac developer of BOINC about this, here's his answer to me:

Quote:

Hi Jord,

This error is one of the great ongoing mysteries of BOINC ;=)
It seems to occur randomly and then disappears by itself. At least, it has always disappeared in every report I've been aware of.

I saw it for a while and added some debugging code, but then the problem disappeared on my system. Thee have been a number of theories about it, but we've never been able to actually pin it down.

Peanut's theory is interesting, but the client state file is written only by the BOINC Client, so I wouldn't expect two cores to be trying to do it at once. It does seem to happen more on Macs than other platforms, but I have heard of it happening on Windows also.

Cheers,
--Charlie

peanut
peanut
Joined: 4 May 07
Posts: 162
Credit: 9,644,812
RAC: 0

I have one other observation

I have one other observation that perhaps Elphidieus could confirm or shoot down.

If Elphidieus gets write state file errors frequently even when tasks are mostly "out of sync" progress wise, perhaps something seemingly unrelated may improve things. When I first started Einstein on my Pro I was getting many write state file errors. Then I ran into the shared memory issue (shmem get failed). I increased the shared memory (which BOINC uses) and the write state file errors seemed to go away for the most part. My Pro still does occasionally get write state file errors, so increasing shared memory does not totally fix the problem. It may not even help at all for sure. I don't know if increasing shared memory did the trick or if it is just better when tasks are out of sync.

If Elphidieus gets errors while tasks are out of sync and has not increased shared memory, then that might lead credence to the idea that increasing shared memory may decrease write state file error frequency.

Elphidieus
Elphidieus
Joined: 20 Feb 05
Posts: 245
Credit: 20,603,702
RAC: 0

RE: I have one other

Message 75493 in response to message 75492

Quote:

I have one other observation that perhaps Elphidieus could confirm or shoot down.

If Elphidieus gets write state file errors frequently even when tasks are mostly "out of sync" progress wise, perhaps something seemingly unrelated may improve things. When I first started Einstein on my Pro I was getting many write state file errors. Then I ran into the shared memory issue (shmem get failed). I increased the shared memory (which BOINC uses) and the write state file errors seemed to go away for the most part. My Pro still does occasionally get write state file errors, so increasing shared memory does not totally fix the problem. It may not even help at all for sure. I don't know if increasing shared memory did the trick or if it is just better when tasks are out of sync.

If Elphidieus gets errors while tasks are out of sync and has not increased shared memory, then that might lead credence to the idea that increasing shared memory may decrease write state file error frequency.

Welp, actually I have increased the amount of shared memory based on what I've researched from the BOINC FAQ Service. Unfortunately that doesn't help.

But anyway, it's a relieve that these error messages does not affect my Mac Pro performance. Thank you all for your input.

Elphidieus
Elphidieus
Joined: 20 Feb 05
Posts: 245
Credit: 20,603,702
RAC: 0

Just to add... I've upped

Just to add...

I've upped my physical memory from 3GB to 5GB, and as of now, eight hours in the running, there hasn't been any error messages from BOINC Manager. Not sure if it's due to the extra RAM to waste....

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.