[lustre-discuss] trimming flash-based external journal device

Andreas Dilger adilger at whamcloud.com
Thu Aug 5 14:23:25 PDT 2021


On Aug 5, 2021, at 13:29, Nathan Dauchy - NOAA Affiliate <nathan.dauchy at noaa.gov<mailto:nathan.dauchy at noaa.gov>> wrote:

Andreas, thanks as always for your insight.  Comments inline...

On Thu, Aug 5, 2021 at 10:48 AM Andreas Dilger <adilger at whamcloud.com<mailto:adilger at whamcloud.com>> wrote:
On Aug 5, 2021, at 09:28, Nathan Dauchy via lustre-discuss <lustre-discuss at lists.lustre.org<mailto:lustre-discuss at lists.lustre.org>> wrote:
Question:  Is it possible that a flash journal device on an ext4 filesystem can reach a point where there are not enough clean blocks to write to, and they can suffer from very degraded write performance?
For the external journal device, this _shouldn't_ happen, in the sense that the writes to this device are pretty much always sequential (except updates to the journal superblock), so as long as there is an erase block that can be cleaned in advance of the next overwrite, it should be essentially "ideal" usage for flash.

I have seen flash devices in the past that are badly behaved and have large spikes in latency when they have to erase blocks when under write pressure.  (this is why periodic fstrim is useful in general)  My concern is that we are dealing with another such device.  It _might_ be the case that the journal device has finally been in service long enough that it has used up all it's clean blocks and now needs to erase "on demand" and hence are performing worse.  (evidence for this is that some newer OSTs added to the same file system do not have the same slowdown)
I know that "fstrim" can be run for mounted ldiskfs file systems, but when I try that it doesn't see the OSTs as using flash, because they are primarily HDD-based.  Is there some other way to tell the system which blocks can be discarded on the journal flash device?  (I found "blkdiscard" but that seems heavyweight and dangerous.)
I don't _think_ you can run fstrim against the journal device directly while it is mounted.  However, you could unmount the filesystem cleanly (which flushes everything from the journal, check no "needs_recovery" feature is set), remove the journal from the filesystem, trim/discard the journal block device, then reformat it as an external journal device again and add it back to the filesystem.

That confirms my understanding.  I may end up going down that path as a good test.  Doesn't sound like much fun though. ;)

Another related question would be how to benchmark the journal device on it's own, particularly write performance, without losing data on an existing file system; similar to the very useful obdfilter-survey tool, but at a lower level.  But I am primarily looking to understand the nuances of flash devices and ldiskfs external journals a bit better.
While the external journal device has an ext4 superblock header for identification (UUID/label), and a feature flag that prevents it from being mounted/used directly, it is not really an ext4 filesystem, just a flat "file".  You'd need to remove it from the main ext4/ldiskfs filesystem, reformat it as ext4 and mount locally, and then run benchmarks (e.g. "dd" would best match the JBD2 workload, or fio if you want random IOPS) against it.  You could do this before/after trim (could use fstrim at this point) to see if it affects the performance or not.

OK, thanks for confirming that there is no magic ext4 journal benchmarking tool.  I'll stop searching.  ;-)

Note that there *are* some journal commit statistics - /proc/fs/jbd2/<dev>/info that you might be able to compare between devices.  Probably the most interesting is "average transaction commit time", which is how long it takes to write the blocks to the journal device after the transaction starts to commit.

Cheers, Andreas
--
Andreas Dilger
Lustre Principal Architect
Whamcloud







-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lustre.org/pipermail/lustre-discuss-lustre.org/attachments/20210805/bc9aba54/attachment.html>


More information about the lustre-discuss mailing list