David Gibbons
2015-May-05 02:19 UTC
[Gluster-users] Geo-Rep. 3.5.3, Missing Files, Incorrect "Files Pending"
So I should do a compare out-of-band from Gluster and see what is actually in-sync vs out of sync? Is there any easy way just to start it over? I am assuming removing and re-adding geo-rep is the easiest way. Is that correct? Thanks, Dave On Mon, May 4, 2015 at 10:09 PM, Aravinda <avishwan at redhat.com> wrote:> Status output has issue showing exact number of files in sync. Please check > the numbers on disk and let us know if difference exists between Master and > Secondary Volume. > > -- > regards > Aravinda > > On 05/05/2015 06:58 AM, David Gibbons wrote: >> >> I am having an issue with geo-replication. There were a number of >> complications when I upgraded to 3.5.3, but geo-replication was (I >> think) working at some point. The volume is accessed via samba using >> vfs_glusterfs. >> >> The main issue is that geo-replication has not been sending updated >> copies of old files to the replicated server. So in the scenario where >> file created -> time passes -> file is modified -> file is saved, the >> new version is not replicated. >> >> Is it possible that one brick is having a geo-rep issue and the others >> are not? Consider this output: >> >> MASTER NODE MASTER VOL MASTER BRICK >> SLAVE STATUS CHECKPOINT STATUS CRAWL >> STATUS FILES SYNCD FILES PENDING BYTES PENDING DELETES >> PENDING FILES SKIPPED >> >> ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ >> gfs-a-1 shares >> /mnt/a-1-shares-brick-1/brick gfs-a-bkp::bkpshares Active >> N/A Hybrid Crawl 2309456 0 0 >> 0 0 >> gfs-a-1 shares >> /mnt/a-1-shares-brick-2/brick gfs-a-bkp::bkpshares Active >> N/A Hybrid Crawl 2315557 0 0 >> 0 0 >> gfs-a-1 shares >> /mnt/a-1-shares-brick-3/brick gfs-a-bkp::bkpshares Active >> N/A Hybrid Crawl 2362884 0 0 >> 0 0 >> gfs-a-1 shares >> /mnt/a-1-shares-brick-4/brick gfs-a-bkp::bkpshares Active >> N/A Hybrid Crawl 2407600 0 0 >> 0 0 >> gfs-a-2 shares >> /mnt/a-2-shares-brick-1/brick gfs-a-bkp::bkpshares Active >> N/A Hybrid Crawl 2409430 0 0 >> 0 0 >> gfs-a-2 shares >> /mnt/a-2-shares-brick-2/brick gfs-a-bkp::bkpshares Active >> N/A Hybrid Crawl 2308969 0 0 >> 0 0 >> gfs-a-2 shares >> /mnt/a-2-shares-brick-3/brick gfs-a-bkp::bkpshares Active >> N/A Hybrid Crawl 2079576 8191 0 >> 0 0 >> gfs-a-2 shares >> /mnt/a-2-shares-brick-4/brick gfs-a-bkp::bkpshares Active >> N/A Hybrid Crawl 2340597 0 0 >> 0 0 >> gfs-a-3 shares >> /mnt/a-3-shares-brick-1/brick gfs-a-bkp::bkpshares Passive >> N/A N/A 0 0 0 >> 0 0 >> gfs-a-3 shares >> /mnt/a-3-shares-brick-2/brick gfs-a-bkp::bkpshares Passive >> N/A N/A 0 0 0 >> 0 0 >> gfs-a-3 shares >> /mnt/a-3-shares-brick-3/brick gfs-a-bkp::bkpshares Passive >> N/A N/A 0 0 0 >> 0 0 >> gfs-a-3 shares >> /mnt/a-3-shares-brick-4/brick gfs-a-bkp::bkpshares Passive >> N/A N/A 0 0 0 >> 0 0 >> gfs-a-4 shares >> /mnt/a-4-shares-brick-1/brick gfs-a-bkp::bkpshares Passive >> N/A N/A 0 0 0 >> 0 0 >> gfs-a-4 shares >> /mnt/a-4-shares-brick-2/brick gfs-a-bkp::bkpshares Passive >> N/A N/A 0 0 0 >> 0 0 >> gfs-a-4 shares >> /mnt/a-4-shares-brick-3/brick gfs-a-bkp::bkpshares Passive >> N/A N/A 0 0 0 >> 0 0 >> gfs-a-4 shares >> /mnt/a-4-shares-brick-4/brick gfs-a-bkp::bkpshares Passive >> N/A N/A 0 0 0 >> 0 0 >> >> This seems to show that there are 8191 files_pending on just one >> brick, and the others are up to date. I am suspicious of the 8191 >> number because it's looks like we're at a bucket-size boundary on the >> backend. I've tried stopping and re-starting the rep session. I've >> also tried changing the change_detector from xsync to changelog. >> Neither seems to have had an effect. >> >> It seems like geo-replication is quite wonky in 3.5.x. Is there light >> at the end of the tunnel, or should I find another solution to >> replicate? >> >> Cheers, >> Dave >> _______________________________________________ >> Gluster-users mailing list >> Gluster-users at gluster.org >> http://www.gluster.org/mailman/listinfo/gluster-users > >
Aravinda
2015-May-05 02:31 UTC
[Gluster-users] Geo-Rep. 3.5.3, Missing Files, Incorrect "Files Pending"
Since we see all status are good either Active or Passive(No Faulty), hoping that everything is in sync(Except the wrong number in status output) find . | wc -l in both Master and Slave mount should help in deciding the number of files in sync. In master nodes, look for log messages. Let us know if you feel any issue in log messages. (/var/log/glusterfs/geo-replication/) In Slave nodes look in /var/log/glusterfs/geo-replication-slaves directory I see, workers are still in Hybrid Crawl state. Please provide the output of gluster volume geo-replication <MASTER> <SLAVEHOST>::<SLAVEVOL> config change_detector Ideally, after initial crawl geo-rep should switch to Changelog crawl. Why it is hard to show exact number of files in Sync? -------------------------------------------------------------------------- Geo-rep doesn't have persistent store of all path names and sync status. When geo-rep gets the list of files to be synced, it adds the number to the counter. But if the same files modified again the counter will be incremented again. Numbers in Status output will not match the number of files on disk. In future we can enhance it by maintaining a db/persistent store to record this information. As of now this is the limitation. -- regards Aravinda On 05/05/2015 07:49 AM, David Gibbons wrote:> So I should do a compare out-of-band from Gluster and see what is > actually in-sync vs out of sync? Is there any easy way just to start > it over? I am assuming removing and re-adding geo-rep is the easiest > way. Is that correct? > > Thanks, > Dave > > On Mon, May 4, 2015 at 10:09 PM, Aravinda <avishwan at redhat.com> wrote: >> Status output has issue showing exact number of files in sync. Please check >> the numbers on disk and let us know if difference exists between Master and >> Secondary Volume. >> >> -- >> regards >> Aravinda >> >> On 05/05/2015 06:58 AM, David Gibbons wrote: >>> I am having an issue with geo-replication. There were a number of >>> complications when I upgraded to 3.5.3, but geo-replication was (I >>> think) working at some point. The volume is accessed via samba using >>> vfs_glusterfs. >>> >>> The main issue is that geo-replication has not been sending updated >>> copies of old files to the replicated server. So in the scenario where >>> file created -> time passes -> file is modified -> file is saved, the >>> new version is not replicated. >>> >>> Is it possible that one brick is having a geo-rep issue and the others >>> are not? Consider this output: >>> >>> MASTER NODE MASTER VOL MASTER BRICK >>> SLAVE STATUS CHECKPOINT STATUS CRAWL >>> STATUS FILES SYNCD FILES PENDING BYTES PENDING DELETES >>> PENDING FILES SKIPPED >>> >>> ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ >>> gfs-a-1 shares >>> /mnt/a-1-shares-brick-1/brick gfs-a-bkp::bkpshares Active >>> N/A Hybrid Crawl 2309456 0 0 >>> 0 0 >>> gfs-a-1 shares >>> /mnt/a-1-shares-brick-2/brick gfs-a-bkp::bkpshares Active >>> N/A Hybrid Crawl 2315557 0 0 >>> 0 0 >>> gfs-a-1 shares >>> /mnt/a-1-shares-brick-3/brick gfs-a-bkp::bkpshares Active >>> N/A Hybrid Crawl 2362884 0 0 >>> 0 0 >>> gfs-a-1 shares >>> /mnt/a-1-shares-brick-4/brick gfs-a-bkp::bkpshares Active >>> N/A Hybrid Crawl 2407600 0 0 >>> 0 0 >>> gfs-a-2 shares >>> /mnt/a-2-shares-brick-1/brick gfs-a-bkp::bkpshares Active >>> N/A Hybrid Crawl 2409430 0 0 >>> 0 0 >>> gfs-a-2 shares >>> /mnt/a-2-shares-brick-2/brick gfs-a-bkp::bkpshares Active >>> N/A Hybrid Crawl 2308969 0 0 >>> 0 0 >>> gfs-a-2 shares >>> /mnt/a-2-shares-brick-3/brick gfs-a-bkp::bkpshares Active >>> N/A Hybrid Crawl 2079576 8191 0 >>> 0 0 >>> gfs-a-2 shares >>> /mnt/a-2-shares-brick-4/brick gfs-a-bkp::bkpshares Active >>> N/A Hybrid Crawl 2340597 0 0 >>> 0 0 >>> gfs-a-3 shares >>> /mnt/a-3-shares-brick-1/brick gfs-a-bkp::bkpshares Passive >>> N/A N/A 0 0 0 >>> 0 0 >>> gfs-a-3 shares >>> /mnt/a-3-shares-brick-2/brick gfs-a-bkp::bkpshares Passive >>> N/A N/A 0 0 0 >>> 0 0 >>> gfs-a-3 shares >>> /mnt/a-3-shares-brick-3/brick gfs-a-bkp::bkpshares Passive >>> N/A N/A 0 0 0 >>> 0 0 >>> gfs-a-3 shares >>> /mnt/a-3-shares-brick-4/brick gfs-a-bkp::bkpshares Passive >>> N/A N/A 0 0 0 >>> 0 0 >>> gfs-a-4 shares >>> /mnt/a-4-shares-brick-1/brick gfs-a-bkp::bkpshares Passive >>> N/A N/A 0 0 0 >>> 0 0 >>> gfs-a-4 shares >>> /mnt/a-4-shares-brick-2/brick gfs-a-bkp::bkpshares Passive >>> N/A N/A 0 0 0 >>> 0 0 >>> gfs-a-4 shares >>> /mnt/a-4-shares-brick-3/brick gfs-a-bkp::bkpshares Passive >>> N/A N/A 0 0 0 >>> 0 0 >>> gfs-a-4 shares >>> /mnt/a-4-shares-brick-4/brick gfs-a-bkp::bkpshares Passive >>> N/A N/A 0 0 0 >>> 0 0 >>> >>> This seems to show that there are 8191 files_pending on just one >>> brick, and the others are up to date. I am suspicious of the 8191 >>> number because it's looks like we're at a bucket-size boundary on the >>> backend. I've tried stopping and re-starting the rep session. I've >>> also tried changing the change_detector from xsync to changelog. >>> Neither seems to have had an effect. >>> >>> It seems like geo-replication is quite wonky in 3.5.x. Is there light >>> at the end of the tunnel, or should I find another solution to >>> replicate? >>> >>> Cheers, >>> Dave >>> _______________________________________________ >>> Gluster-users mailing list >>> Gluster-users at gluster.org >>> http://www.gluster.org/mailman/listinfo/gluster-users >>