Make the code stop if IO fails?

[ykempf]

Do we want the code to stop if IO fails eg due to disk quota being full? Or do we want to just chug along in the hope that someone notices and cleans up? #1Mburn

[lkotipal]

I'd say bulk write fails qualify for fatal errors since we'll end up with a long stretch of missing files if someone isn't there just in time to catch the error. However, restart writes tend to fail more often (due to the size) and might be recoverable; if the bad restart file is cleaned up and the user notified, it's reasonable to catch that in the output before the quota gets so full nothing can be written.