How long does it take to pause a work unit?

[TiO2]

I've been pausing work units before shutting down to try to avoid losing work units, but I notice that if I go shut down right after hitting pause I will often still lose the work unit (even though the interface shows it as paused). How long does it take to actually pause a WU, and how long should I wait until I actually shut down the computer to avoid losing work?

I just lost a 95.6% completed work unit with over 8 hours of work from hitting shut down right after hitting pause (as I was about to leave the house), which is particularly infuriating:

[TiO2]

Also why is this all so poorly documented? Not everyone can afford to fold 24/7; electricity is expensive where I live and I can only justify running my rig when I’m occupying the room since I need to heat the room anyways. I need to often shut down the rig when I leave.

Why doesn’t the V8.4 client guide mention work unit corruption on shutdown at all?

[Joe_H]

In my experience a CPU WU usually takes under 30 seconds to pause. That includes writing out a checkpoint. GPU folding cores are a bit different. They have processing loops interfacing with the video driver that can only be interrupted at certain points. So give them a minute or so.

For an example of a CPU WU being paused:

Code: Select all

18:23:15:I1:WU6834:Completed 7500 out of 250000 steps (3%)
18:24:45:I1:Machine state pause
18:24:45:I1:WU6834:Caught signal SIGINT(2) on PID 55298
18:24:45:I1:WU6834:Exiting, please wait. . .
18:25:13:I1:WU6834:Folding@home Core Shutdown: INTERRUPTED
18:25:13:I1:WU6834:Core returned INTERRUPTED (102)

or 28 seconds from the time the Pause signal was sent until the final exit was processed and logged back to the client.

[TiO2]

Thanks for the info. If pausing and waiting for a minute is a hard requirement for Windows users when shutting down to avoid losing all their work, why isn’t any of this information in the client guide or somewhere where users can easily find it (such as in the client itself in bright red letters)?

[Joe_H]

Why, because the developer who wrote the guide hasn't gone back in to document the one exception that shouldn't be needed. The client was programmed to MS's guidelines to follow and work with shutdown. But Windows Shutdown still ignores the wait from the client and just keeps on going and shuts down not cleanly. The v8.5 public beta has additional code that should improve this, but people haven't reported on whether it works or not.

The client does shutdown cleanly for Linux and macOS. That it gets ignored by Windows Shutdown has been posted frequently here and on the discord.

[TiO2]

It’s basically irrelevant that the bug is Microsoft’s fault; if it’s there it’s on the FAH project to clearly document it and any required workarounds. This is a fundamental, unfixed bug that throws away huge amounts of donors’ work and affects basically every Windows donor; is it such a big ask to at least clearly document it and make sure donors know to pause and wait before shutdown, without having to go into forums/Discord, which most people won’t do? It reflects really poorly on the project.

[muziqaz]

You are mentioning shutting down right after you pause. How "right after" is it? ns after? A second after? Are you holding shutdown button/icon ransom, while you are pausing folding? The reason I ask, is that your issues are not widespread and are extremely rare, and specific to your ability to shutdown your systems in microseconds. The rest of the folding world do not have your skills, including testers, and the dev. The reason this corner case is not documented anywhere, is because none of us encountered it. You see, when we pause (we don't need to anymore, btw, since latest beta fixed the issue with Windows killing folding WU upon system restart) our folding activities, we shutdown the system like normal human beings, you know, move the mouse to start button, then Power button, then select what we want to do (restart, suspend, shutdown). By the time we are at Shutdown stage, fahclient v8 already have saved pause state (old v7 client used to need 30 seconds or so to save fold state). Even if you do cool kids way, by clicking Windows key on keyboard and then navigating to shutdown, that is still enough time for client to record pause state. Now, if you are still using HDD from 1998 (and SDRAM as system memory), then yes, FAHclient will have to wait for HDD to write things, but I really doubt this is the case with you. Any SATA SSD or more modern SSDs, will have no issues with fahclient. Same with modern DDR memory.

There are a lot of things which reflect badly on the project, this one isn't it

[TiO2]

I hit pause, then move the cursor over to the start menu and hit "Shutdown", which takes about a second or two. I don't see why those assumptions were necessary.

It's good that the bug is supposedly fixed in the 8.5 beta, but most people are not running the beta. I run 8.4.9 because that's the "stable" public release that's on the FAH website, and it's not documented there at all. I've lost at least 2 mostly-finished GPU WUs due to not waiting after pausing, and at least another 3 WUs before I knew I needed to pause at all before shutdown/standby. There are at least four other recent forum posts where donors encountered this bug, so it seems to be a widespread issue - one which could have been avoided with basic documentation.

[Joe_H]

I take it you are volunteering to update the installation guide? Or are you expecting someone else to do it? The reality is that much of the support is done by or was done by volunteers. The paid developer did include writing up a guide for v8 installs. It could be improved, but their concentration right now is on getting a stable public release of the v8.5 client out. Then they will be working on an updated CPU folding core to utilize recent updates to the GROMACS code library. The volunteer who had maintained those installation guide pages up to version 7 passed away just before COVID and in the middle of the F@h project moving out of Dr. Pande's Stanford lab. No permanent replacement so far that I know of.

Other than posting beta tests of WUs most researchers and the developer do not hang out here. You are welcome to send your complaints or offers of volunteering to them, I understand part of the discord is there for volunteers. You can even post on the GitHub for the F@h client and web interface.

[muziqaz]

I hit pause, then move the cursor over to the start menu and hit "Shutdown", which takes about a second or two. I don't see why those assumptions were necessary.

It's good that the bug is supposedly fixed in the 8.5 beta, but most people are not running the beta. I run 8.4.9 because that's the "stable" public release that's on the FAH website, and it's not documented there at all. I've lost at least 2 mostly-finished GPU WUs due to not waiting after pausing, and at least another 3 WUs before I knew I needed to pause at all before shutdown/standby. There are at least four other recent forum posts where donors encountered this bug, so it seems to be a widespread issue - one which could have been avoided with basic documentation.

Your linked threads are not talking about your issue. Your issue is apparently that client does not save the pause status before you hit shutdown. Issue in your linked threads was an issue we had since the beginning of time, due to boneheads at MS. Now we have that issue fixed in latest beta. Documentation only goes so far. We have discord, forum, various team's forums. Besides, you are free to request a better documentation in github, where developer actually actively participates in discussion. Here, you are preaching to the choir

[TiO2]

The bug and its workaround can be documented in just one sentence:

Windows users: If you need to shut down your computer or put it to sleep, please pause all work units and wait at least a minute before doing so, or else you may lose all progress on a work unit due to a bug with Window's shutdown process.

That's all you need, and could have saved a lot of dumped work units and the associated heartache.