Page MenuHomePhabricator

20241201 wikidatawiki xml dump not progressing
Closed, ResolvedPublic

Description

We've been seeing email alerts like:

[data engineering alerts] [ops-dumps] PASS: dump-jobs-watcher
Systemd timer ran the following command:

    /bin/bash /usr/local/bin/job_watcher.sh --dumpsbasedir /data/xmldatadumps/public --locksbasedir /data/xmldatadumps/private

Its return value was 0 and emitted the following output:

PROBLEM: wikidatawiki has file wikidatawiki/20241201/wikidatawiki-20241201-pages-meta-history9.xml-p12828996p12875404.bz2.inprog at least 4 hours older than lock
...

This looks similar to the symptom reported in T376926: frwiki trashing for 20241001 dump.

Event Timeline

Ottomata removed a project: Epic.
Ottomata added subscribers: xcollazo, Milimetric.

@xcollazo recommended we try to kill (and restart?) the job, as described in T376926#10218853.

I'll try that now...

Mentioned in SAL (#wikimedia-operations) [2024-12-12T17:53:57Z] <ottomata> killing wikidatawiki xml dump process to try to unstick it - T382084

17:45:21 [@snapshot1011:/home/otto] $ cd /mnt/dumpsdata/xmldatadumps/private/wikidatawiki
17:45:23 [@snapshot1011:/mnt/dumpsdata/xmldatadumps/private/wikidatawiki] $ cat lock_20241201
snapshot1011.eqiad.wmnet 3510981

17:50:49 [@snapshot1011:/mnt/dumpsdata/xmldatadumps/private/wikidatawiki] $ sudo -u dumpsgen  bash
dumpsgen@snapshot1011:/mnt/dumpsdata/xmldatadumps/private/wikidatawiki$ cd /srv/deployment/dumps/dumps/xmldumps-backup

dumpsgen@snapshot1011:/srv/deployment/dumps/dumps/xmldumps-backup$ python3 dumpadmin.py --kill --configfile /etc/dumps/confs/wikidump.conf.dumps:bigwikis --wiki wikidatawiki --dryrun
would kill processes ['418228', '418229', '418230', '418236', '418241', '418242', '418243', '418244', '418245', '418246', '418247', '418259', '418262', '418265', '418701', '418732', '418810', '418823', '418826', '418862', '418901', '418906', '419048', '419094', '419136', '419164', '419862', '422623']

python3 dumpadmin.py --kill --configfile /etc/dumps/confs/wikidump.conf.dumps:bigwikis --wiki wikidatawiki

dumpsgen@snapshot1011:/srv/deployment/dumps/dumps/xmldumps-backup$ ls -l /mnt/dumpsdata/xmldatadumps/private/wikidatawiki/lock_20241201
-rwxr-xr-x 1 dumpsgen dumpsgen 32 Dec 12 17:54 /mnt/dumpsdata/xmldatadumps/private/wikidatawiki/lock_20241201

dumpsgen@snapshot1011:/srv/deployment/dumps/dumps/xmldumps-backup$ rm /mnt/dumpsdata/xmldatadumps/private/wikidatawiki/lock_20241201

Done. But hm, maybe we should remove the lockfile before killing?

Discussed with Xabriel in slack. I was worried about the fact that there is no lockfile, so I killed the processes again.

dumpsgen@snapshot1011:/srv/deployment/dumps/dumps/xmldumps-backup$ python3 dumpadmin.py --kill --configfile /etc/dumps/confs/wikidump.conf.dumps:bigwikis --wiki wikidatawiki --dryrun
would kill processes ['432994', '432995', '432996', '432997', '432998', '432999', '433000', '433001', '433002', '433003', '433004', '433005', '433006', '433007', '433008', '433009', '433010', '433011', '433012', '433013', '433014', '433015', '433018', '433021', '433023', '433026', '433028', '433204', '433290', '433320', '433428', '433731', '433754', '433757', '433760', '433765', '433827', '433839', '433847', '433864', '433894', '433960', '434259', '434328', '434430', '434549', '434556', '438970', '442852', '443110', '444439', '447923']

dumpsgen@snapshot1011:/srv/deployment/dumps/dumps/xmldumps-backup$ python3 dumpadmin.py --kill --configfile /etc/dumps/confs/wikidump.conf.dumps:bigwikis --wiki wikidatawiki

Looks like last logs from wikidatawiki 2024-12-01 run are from 2024-12-12:

xcollazo@snapshot1011:/mnt/dumpsdata/xmldatadumps/private/wikidatawiki/20241201$ pwd
/mnt/dumpsdata/xmldatadumps/private/wikidatawiki/20241201
xcollazo@snapshot1011:/mnt/dumpsdata/xmldatadumps/private/wikidatawiki/20241201$ tail -n 5 dumplog.txt 
2024-12-12 19:30:09: wikidatawiki (ID 433008) 5993 pages (1.0|314.8/sec all|curr), 280000 revs (48.9|52.5/sec all|curr), 99.8%|99.8% prefetched (all|curr), ETA 2026-06-07 11:26:03 [max 2286480581]
2024-12-12 19:30:09: wikidatawiki (ID 432995) 18901 pages (3.3|566.2/sec all|curr), 477000 revs (83.2|30.0/sec all|curr), 99.9%|99.9% prefetched (all|curr), ETA 2025-10-26 17:06:08 [max 2286480585]
2024-12-12 19:30:09: wikidatawiki (ID 433005) 10904 pages (1.9|1350.5/sec all|curr), 360000 revs (62.8|123.9/sec all|curr), 99.8%|99.8% prefetched (all|curr), ETA 2026-02-07 03:12:26 [max 2286480581]
2024-12-12 19:30:10: wikidatawiki (ID 433003) 11039 pages (1.9|1047.1/sec all|curr), 395000 revs (68.9|94.9/sec all|curr), 99.9%|99.9% prefetched (all|curr), ETA 2025-12-31 18:43:54 [max 2286480585]
2024-12-12 19:30:14: wikidatawiki (ID 433000) 3466 pages (0.6|152.7/sec all|curr), 243000 revs (42.4|44.0/sec all|curr), 99.9%|99.9% prefetched (all|curr), ETA 2026-08-29 12:18:40 [max 2286480584]

So the process killing did not re-start the dumps.

Let's try and finish it. snapshot1011 will soon be busy with the 2024-12-20 partial run, so let's run it on snapshot1014:

sudo -u dumpsgen bash
cd /srv/deployment/dumps/dumps/xmldumps-backup

screen -S xcollazo-wikidatawiki-20241201-run

bash ./worker --date 20241201 --skipdone --exclusive --log --configfile /etc/dumps/confs/wikidump.conf.dumps:bigwikis --wiki wikidatawiki

If we finish successfully, we'd need to force rsync to the clouddumps. @Milimetric do we have how to force rsync documented somewhere?

@xcollazo can we consider this resolved?

We still need to force rsync as per T382084#10418942.

@Milimetric could you share the steps?

Ottomata claimed this task.

According to @Milimetric this is resolved.