[LU-4258] ENOSPC when using conf_param on fs with a lot of OSTs

• Type: Bug
• Resolution: Cannot Reproduce
• Priority: Minor
• Affects Version/s: Lustre 2.4.1
• Severity: 3
• Assignee: Oleg Drokin
• Reporter: Manish Patel (Inactive)

Description

At TACC we are running into an issue where the config log is overflowing due to the large number of OSTs and conf_param on the mdt and clients is failing with ENOSPC. I tracked down the issue to llog_osd_write_rec:

00000040:00000001:3.0:1384532145.376864:0:12518:0:(llog_osd.c:336:llog_osd_write_rec()) Process entered
00000040:00001000:3.0:1384532145.376865:0:12518:0:(llog_osd.c:346:llog_osd_write_rec()) new record 10620000 to [0xa:0x419:0x0]
00000040:00000001:3.0:1384532145.376867:0:12518:0:(llog_osd.c:440:llog_osd_write_rec()) Process leaving (rc=18446744073709551588 : -28 : ffffffffffffffe4)

It looks like it's hitting this:

    /* if it's the last idx in log file, then return -ENOSPC */                     
    if (loghandle->lgh_last_idx >= LLOG_BITMAP_SIZE(llh) - 1)                       
        RETURN(-ENOSPC);                                                            

Here is the CONFIGs directory for reference:

total 31212
-rw-r--r-- 1 root root 11674048 Sep 24 13:39 gsfs-client
-rw-r--r-- 1 root root 11980704 Sep 24 13:39 gsfs-MDT0000
-rw-r--r-- 1 root root     9432 Sep 24 13:48 gsfs-OST0000
-rw-r--r-- 1 root root     9432 Sep 24 13:49 gsfs-OST0001
...
-rw-r--r-- 1 root root     9432 Oct  1 12:15 gsfs-OST029f
-rw-r--r-- 1 root root    12288 Sep 24 13:39 mountdata

Is there a way to increase the BITMAP_SIZE? It looks like the bitmap itself is based on the CHUNK_SIZE but the BITMAP_SIZE macro doesn't reference it at all.

Thanks.