Did something recently changed in ignores handling ?

#1

I met deleted files in a folder since FS Watcher.

One out of three devices has !ignores when the 2 other have not:

!201402 foo
2014/*
!2014
!200604 07 foo
!200604 17-23 bar
2006/*
!2006
*

Unintended deletion were on the devices that don’t have !ignores. One of them is up 24/24.

I work this way : I edit files in say 2006/2006 07 foo in the device having the !ignores. I wait/check for sync to the 24h up device, then I remove the !200604 07 foo line out of the .stignore file with the GUI, then I delete the local ..2006/2006 07 foo directory.

I think the data loss happened on the 24 up device, as I found the files back in the staggered trash can with time stamps before the second device was fired up.

Same thing happened in other directories I recently worked on.

(Audrius Butkevicius) #2

Yes there was, unignores now traverse deeper, not needing to unignore directories along the way.

#3

Thank you Audrius. Please when was this change introduced ? 47? 48? 49? rc?

Was there anything changed in the documentation about this, or was it/still is it undocumented ? I just had look but was unable to remember old doc if ever.

(David) #4

Have a look at this PR, went into v0.14.49-rc.1.

(Audrius Butkevicius) #5

Summoning @imsodin. I guess we’ll need a repro.

#6

I get the following issue, maybe because releases mismatch:

0.14.49-rc1:

!/2012/2012 01 foo
*

0.14.48:

no ignore, no !ignore

Subtree exists at both end with 138 files. When added a file at 49-rc1 side, 48 stable showed 139 items Out of Sync

(Simon) #7

I reproduced observed the out of sync issue and know how it happens - it is due to the new need database stuff in combination with ignored files, but not related to descending to catch include paths. It’s totally possible that what you describe in your first post is unrelated to this (haven’t found anything about that yet).

(Simon) #8

I can’t reproduce the unintended deletions. Please check my steps and correct me if I did something different:

  1. Setup share between devices A and B.
  2. Add the exact ignore patternss of your first post on device A.
  3. Create folder 2006/200604 07 foo and some random files therein on A and wait for sync to B.
  4. Remove !200604 07 foo from the ignore patterns on A.
    Intermediate observation: Local state goes down to just 1 folder (2006).
  5. On A delete 200604 07 foo inside of 2006 (i.e. I don’t delete 2006 itself).

My result: Nothing happens on B.

Your result (?): The files are also deleted on B.

#9

Hi Simon

Differences from me in your steps:

Everything files/dirs/shares/!ignores yet existed at the moment ST was upgraded on A (the !ignoring device) was upgraded to 0.14.49-rc1. The 2 other devices in the cluster are/was set to autoupgrade to stable releases only. At the time of upgrade, the cluster was UpToDate on the 3 everywhere.

I can’t remember if I deleted 200604 07 foo then the related !ignore line before or after the upgrade.

Like you I didn’t deleted the 2006 directory as I still have the !200604 17-23 bar to work on inside.

Is this is the new best syntax for this goal?:

!/2006/200604 07 foo
!/2006/200604 17-23 bar
!/2014/201402 foo
*

Notice the leading / I added to feel secure

Also, the strange thing, before & after I reset-database on the 3 nodes is that B & C show(resurects) missing OutOfSync items that where deleted several months ago.

(Russell) #10

See Deleted files remain ‘Syncing (95%, 0B)’ for report of resurrected ‘zombie’ files after upgrade to .49rc. I was experimenting with ignores on separate folder when zombies (deleted files from days ago) appeared as out-of-sync. Reverting to .48 made no difference. Resolved by --reset-deltas on offending machine and deleting ignores. Updated again to .49rc and no further problem but yet to reintroduce ignores. I might sit on my hands for a while.

#11

[EDIT] other things: I found fswatcher was enabled on dev C with default max_user_watches (8192) which was roughly 1/4 the number of files in all the shares on this dev. I disabled and set MUW to 131072 for further re-enablement.

(Simon) #12

Don’t worry about the out of sync stuff, it is most likely https://github.com/syncthing/syncthing/issues/5007 for which a PR with a fix is already in place - there will be an RC 2 soon.

And yes, those ignore patterns should work with 0.14.49.

As to reproducing the deletion issue: I don’t see anything to try/change on my reproducer based on your answer (or I didn’t understand). If you can reproduce it, please post steps the way I did. As it stands there is nothing more that I can investigate about this for me.

#13

Thank you guys. I was so puzzled by these issues that I didn’t notice today the Out Of Sync stuff went away.

Have a good sunday

(Audrius Butkevicius) #14

It would still be nice if you produced steps for the deletion reproduction issue.

#15

I’ll do my best Audrius, but it will be very difficult as I didn’t change the way I usually work with the files in this shared folder. As I said, given the !ignored syntax I used while on v48, maybe the introduction of new behaviour in 49rc did something, maybe plus a bad setting in one 48stable about fsWatcher. Now I changed it on my 49rc device.

(Audrius Butkevicius) #16

Ignores and watcher is not related.

#17

Maybe also related to #5007 : device A is filling its own Recent Changes GUI list each minute, as well as the 2 other devices one, with:

Device Action Type Folder Path Time
DevA modified dir Photos 2012 2018-06-17 15:46:55
DevA modified dir Photos 2012 2018-06-17 15:46:08
DevA modified dir Photos 2012 2018-06-17 15:45:11
DevA modified dir Photos 2012 2018-06-17 15:44:18
DevA modified dir Photos 2012 2018-06-17 15:43:12

Although all 3 devices show UpToDate, B & C blink each minute showing Synching 0% (intended as fsWatcher is disabled), and sometimes briefly Out Of Sync 99% missing 1 item for 128B.

[EDIT] I commented out all !ignores but one (!/2012/201208 foo … and * of course :wink: )

(Audrius Butkevicius) #18

I don’t think it it should be related to the rc

#19

Now I uncommented !/2009/200912 26* and the Recent Changes lists went:

Device Action Type Folder Path Time
DevA modified dir Photos 2012 2018-06-17 18:33:02
…
DevA modified dir Photos 2009 2018-06-17 18:24:52
DevA modified dir Photos 2012 2018-06-17 18:23:46
DevA modified dir Photos 2009 2018-06-17 18:23:46
DevA modified dir Photos 2012 2018-06-17 18:22:45
DevA added dir Photos 2009 2018-06-17 18:22:45
…
#20

… and went on my usual work by treating my local .../2009/200912 26 etc end of year. The remaining job was to look at the pictures inside, decide which ones to keep and delete the others. Before I started I checked the Up to Date status on the 3 devices (OK) and I cleaned .stversions/2009 everywhere to track any further possible issue.

I did the job, then locally added added my usual flag inside the subdir (a 0 byte empty file called 0k), then I was ready to remove the !ignore and delete the local subdir (and even the no more needed 2009 dir as there is the single 200912blah dir inside) after I check everything is still UptoDate … but:

Now, devices B & C show syncthing 99% ~265MiB to Device A for 107 items! Local subdir is 123 items ~297MB, remote ones (B & C) are 153 items ~382MB, and remote .stversions/2009 are only 13 files ~20MB when I deleted 43 (153+13-123) on dev A.

To have better/cleaner view I re-commented !2012/... so that only one !ignore remains, the 2009 one.

On A Local State shows 123 files/1 dir (what maybe ok or ko, depending we count 2009 for one, and 200912 26... for an other)

While A is scanning, Global State dir number decreases by one (1) and increases back by 1 in the end, on every scan even no real change happened.

The 107 files B & C would want to send to A either are & were yet on A, or are neither in A nor in B/C’s .stversions .

Either initial Up to Date was wrong (actually it wasn’t, I checked this) or st was puzzled at some point during the deletions on devA.