Re: kern/58776: RAIDframe panic on I/O error during reconstruction

["oster%netbsd.org@localhost via gnats" <gnats-admin%NetBSD.org@localhost>]

The following reply was made to PR kern/58776; it has been noted by GNATS.

From: oster%netbsd.org@localhost
To: gnats-bugs%netbsd.org@localhost
Cc: 
Subject: Re: kern/58776: RAIDframe panic on I/O error during reconstruction
Date: Wed, 30 Oct 2024 19:28:52 +0000

 October 28, 2024 at 7:30 AM, "Emmanuel Dreyfus via gnats" <gnats-admin@ne=
 tbsd.org> wrote:
 
 >=20=20
 >=20 This time it was wd2 write error=20
 [snip]
 >=20 [ 191991.0970825] raid1: Recon write failed (status 5(0x5))!
 >  [ 191991.0970825] raid1: reconstruction failed.
 >  [ 192001.1003051] wd2d: device timeout writing fsbn 15966377986 of 159=
 66377986-15966378017 (wd2 bn 15966377986; cn 15839660 tn 11 sn 13)
 >  [ 192001.1122002] wd2d: error writing fsbn 15966377986 of 15966377986-=
 15966378017 (wd2 bn 15966377986; cn 15839660 tn 11 sn 13)
 >  [ 192001.1220854] raid1: Recon write failed (status 5(0x5))!
 >  [ 192001.1220854] raid1: 566502314 recon event waits, 11 recon delays
 >  [ 192001.1303189] raid1: 2821808363 max exec ticks
 >  [ 192011.1253150] wd2d: device timeout writing fsbn 15966378018 of 159=
 66378018-15966378049 (wd2 bn 15966378018; cn 15839660 tn 11 sn 45)
 >  [ 192011.1372114] wd2d: error writing fsbn 15966378018 of 15966378018-=
 15966378049 (wd2 bn 15966378018; cn 15839660 tn 11 sn 45)
 >  [ 192011.1470958] raid1: Recon write failed (status 5(0x5))!
 
 Thanks for this.
 
 So this "Recon write failed" is showing up *after* raid1 thinks that=20
 the=20reconstruction has failed and is already done...  The code in=20
 rf_reconstruct.c:ProcessReconEvent()=20in the RF_REVENT_WRITE_FAILED=20
 case=20may not be sufficient.  More likely, it's the code in=20
 rf_reconstruct.c:rf_ContinueReconstructFailedDisk()=20in the
 'if (recon_error) { /* we've encountered an error in reconstructing. */` =
 case,=20
 where=20perhaps we're not waiting for enough writes to complete?=20=20
 (I=20need to look at how the IOs are scheduled again to figure out=20
 if/how=20those 'extra writes' are getting generated, and then exactly=20
 how=20to account for them..)
 
 In any event, this error path is not as well tested as it could/should be=
 .
 
 Later...
 
 Greg Oster