FTP service slow and timing out when writing to GFS2 filesystem in Active/Active mode Red Hat High Availability cluster
Issue
-
We need help to limit memory for filesystem cache. The symptom is that application is intermittent slowness to accessing files in loader, or FTP.
- Many tasks are showing "blocked for more than 120 seconds" in /var/log/messages:
kernel: INFO: task intransporter:31104 blocked for more than 120 seconds. kernel: INFO: task intransporter:31104 blocked for more than 120 seconds. kernel: INFO: task pdflush:3628 blocked for more than 120 seconds. kernel: INFO: task intransporter:31104 blocked for more than 120 seconds. kernel: INFO: task disposer:31102 blocked for more than 120 seconds. kernel: INFO: task intransporter:31104 blocked for more than 120 seconds. kernel: INFO: task intransporter:31104 blocked for more than 120 seconds. kernel: INFO: task intransporter:31104 blocked for more than 120 seconds. kernel: INFO: task intransporter:17053 blocked for more than 120 seconds. kernel: INFO: task pdflush:3628 blocked for more than 120 seconds.
Environment
-
Red Hat Enterprise Linux 5.8 (RHEL5.8)
- Red Hat Enterprise Linux Server 5 (with the High Availability Add on)
- Red Hat Enterprise Linux Server 6 (with the High Availability Add on)
- Red Hat High Availability cluster with two or more nodes
-
GFS2 filesystem
- Filesystem mounted on 2 or more nodes simultaneously
- GFS2 filesystem performance is slow/lower than expected.
- Plock rate limit is set to unlimited in /etc/cluster/cluster.conf:
<gfs_controld plock_rate_limit="0"/> <dlm plock_ownership="1" plock_rate_limit="0"/>
- Majority of file access is to 2 specific directories (files created and removed simultaneously by multiple cluster nodes).
-
Workload is described as "Active/Active write access" from multiple cluster nodes.
- Very I/O intensive workload
- Many clients connect to all and create files on the same GFS2 filesystem in the same folder.
- An application moves files from ftp incoming directory to processing directory on the same GFS2 filesystem.
- A second application processes the file, but does not change its contents.
- A third application determines when a file has finished processing and cleans it up.
- Until the file has been processed and cleaned up, clients poll both shared directories every 10 seconds to determine if processing of that file has finished yet.
- FTP filesystem access (if FTP is used) is sometimes slow - (200mb/200s) ftp performance both reading and writing to gfs2
- There are no backups configured for this filesystem.
- Sar data doesn't show excessive load when the issue occurs. Constant moderate load throughout the day.
- Issue began after increased load was applied to GFS2 filesystem and clustered application.
Subscriber exclusive content
A Red Hat subscription provides unlimited access to our knowledgebase, tools, and much more.