[plug] Data Loss with Ext4, SSD & a Power Outage

[Tim]

Just hoping someone here can shed light on this one.

My /home is mounted on a SSD, and I was working on a new file the other day
(source code), using PHPStorm as my editor. (I'm normally a Vim user, but
having been trialling it for some of it's source code refactoring tools).
I had saved the file a number of times, and know the saves had been
"working" as I'd been loading the PHP Class via the webserver, testing the
components I was working on.
Then the power "failed" (a child bumped the powerpoint), and when I
restarted the computer, the file I'd been working on, and the workspace.xml
file PHPStorm uses to keep track of open files, were both empty. They both
existed, but it was if they were truncated.
Given I'd only lost a small amount of work (it was a new class file), I
started it again. 10 minutes later, the same happened, a child bumped the
power cable, upon restart the same 2 files were again "truncated".
This time however, I'd gotten a bit more work done, and know I'd hit save
many times while I'd worked, and checked both my Dropbox and my local
script rdiff-backup script, and sure enough the file contents were in both
locations. After the restart, Dropbox had of course "truncated" the file on
the server to match my client, but thanks to revisions I was able to
rollback to my last saved revision, and only lost 1 line, the line I was
typing before the power outage.

So given that both Dropbox and my rdiff-backup script had the file
contents, why was the file being truncated on the SSD, the origin for the
file?

I can only assume that a file write cache was to blame, but my reading of
'man mount' suggests that the commit interval is 5 seconds. And that's
supposed to be metadata and data. I checked with a few mount commands and
verified that if you set the commit interval to the default of 5, it
doesn't show in the mount options, changing it from the default you then
see it in the mount options.

$ cat /proc/mounts |grep home
/dev/sdg2 /home ext4 rw,relatime,errors=remount-ro,data=ordered 0 0

The destination for the rdiff-backup script is local, but on a spinning
disk HDD, and it's copy was correct. It's mount options are:
/dev/sde3 /mnt/data1 ext4 rw,relatime,data=ordered 0 0

The only other thing I can think of, is that the PHPStorm editor doesn't
send a Sync command, allowing those writes to sit in cache, while
rdiff-backup sends a Sync. But if that's the case, then the commit interval
that mount talks about isn't accurate, and my googling pointing me to the
commit interval is also wrong.

Any suggestions? I'm willing to recreate the scenario to try and work out
why it's happening.

Tim

-- 
Timothy White - Somewhere in Australia
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.plug.org.au/pipermail/plug/attachments/20140902/28fa69e0/attachment.html>