Re: network errors

"Also sprach [email protected]:"
[Charset iso-8859-1 unsupported, filtering to ASCII...]
> >> Is there any option to make the ENBD client less sensitive to network
> >> errors? I am using the -e option, but I don't want to pass temporary
> >> network failures (like a small period of high network trafic) to the RAID
> >> driver.
> 
> >If you are using FR1 instead of raid1, it retries read errors, and
> >resyncs the disk intelligently and rapidly if necessary.
> 
> I can't use FR1, my ENBD client also has root raid, so to use the FR1 I
> have to patch the kernel or boot with a initrd. And this is a production
> system, I don't know how much stable is the FR1 patch....

I don't think anyone has stats on anything. 

Are you saying that you want to use kernel raid1 for one md device and
fr1 for another?

You know, I think I can arrange that. I'd just have to change the
ifdef's in the code for a dynamic test one on a flag set on the device
itelf.

Would you like me to do that?


> >Hmm. Depends on the kind of error! If show_errs is not set, then errors
> >such as network timeouts will always be retried in the enbd driver and
> >never passed on to raid.
> 
> >If you like I can add a count to limit the number of retries.
> 
> Yes, I think this will solve the problem. Whithout show_errs, the system
> will be less sensitive to temporary errors, but with a limitation in the
> number of retries it will eventualy show errors caused by more serious
> failures. Can you add this as an option to enbd-client?

OK. I'll make a try when it cools down outside tomorrow morning ... 

Peter