PG: primary should not be in the peer_info, skip if it is

[neha-ojha]

Skip the primary when iterating over peer_info to avoid adding the primary twice.

Signed-off-by: Neha Ojha nojha@redhat.com

[dzafman]

@neha-ojha I'm enhancing this function right now. Actually, the primary is never in peer_info, so I'm not clear this is needed.

Did you see this happen or was this from code inspection?

[neha-ojha]

@dzafman I hit this assert (target >= missing_target_objects.size()) http://qa-proxy.ceph.com/teuthology/nojha-2018-01-29_22:11:59-rados-wip-async-recovery-2018-01-29---basic-smithi/2124263/teuthology.log

Adding some debug logs showed the following:

2018-01-29 22:25:12.812 7fb3b160a700 10 osd.2 pg_epoch: 27 pg[2.e( v 27'1271 lc 17'2 (0'0,27'1271] local-lis/les=26/27 n=1267 ec=12/12 lis/c 26/16 les/c/f 27/17/0 26/26/26) [2,0] r=0 lpr=26 pi=[12,26)/2 luod=27'1270 crt=27'1271 mlcod 17'2 active+recovery_unfound+degraded m=2 u=2] _update_calc_stats target = 2

2018-01-29 22:25:12.812 7fb3b160a700 10 osd.2 pg_epoch: 27 pg[2.e( v 27'1271 lc 17'2 (0'0,27'1271] local-lis/les=26/27 n=1267 ec=12/12 lis/c 26/16 les/c/f 27/17/0 26/26/26) [2,0] r=0 lpr=26 pi=[12,26)/2 luod=27'1270 crt=27'1271 mlcod 17'2 active+recovery_unfound+degraded m=2 u=2] _update_calc_stats missing_target_objects size = 3

2018-01-29 22:25:12.812 7fb3b160a700 10 osd.2 pg_epoch: 27 pg[2.e( v 27'1271 lc 17'2 (0'0,27'1271] local-lis/les=26/27 n=1267 ec=12/12 lis/c 26/16 les/c/f 27/17/0 26/26/26) [2,0] r=0 lpr=26 pi=[12,26)/2 luod=27'1270 crt=27'1271 mlcod 17'2 active+recovery_unfound+degraded m=2 u=2] _update_calc_stats missing_target_objects = 0,2,2,0,2,2

2 being the primary here was getting added twice.

[dzafman]

Let's add a comment that this shouldn't be needed because the primary should be in the peer_info.

[dzafman]

@neha-ojha Also, change the commit comment to reflect that the primary isn't supposed to be in peer_info.

[jdurgin]

@dzafman we've got a bunch of loops over peer_* data structures that explicitly skip the primary - it does seem like a bad pattern, but they seem to be preventing bugs for now.

[dzafman]

@jdurgin Yes, I talked with @neha-ojha about it. I created a tracker to figure out later why primary is ending up in peer_info. http://tracker.ceph.com/issues/22834

[tchaikov]

• http://pulpito.ceph.com/kchai-2018-02-01_09:17:08-rados-wip-kefu-testing-2018-02-01-1258-distro-basic-mira/
• http://pulpito.ceph.com/kchai-2018-02-02_05:30:30-rados-wip-kefu-testing-2018-02-01-1258-distro-basic-mira/