Red Hat Bugzilla – Bug 1299864
Incorrect results from "Edit Release Components"
Last modified: 2017-01-12 23:57:00 EST
Description of problem:
I've started seeing some flakiness in the "edit release components" screen. I'm a frequent user of this functionality for RHEL6 and RHEL7 releases (currently working on 6.8), and I have recently seen 2 incorrect results.
1) I added a new component to "current approved components", it told me it was added, but it wasn't there.
2) I removed a component from the "current approved components", i got the confirmation mail that it was removed. Then I removed 4 more components with a single "save approved", and the confirmation message told me that it removed the 4 components AND that it added back the one I had removed in the previous action (which I did NOT do).
Version-Release number of selected component (if applicable):
Steps to Reproduce:
I can't say with 100% certainty that this isn't a user error, so I'm closing the bug. The info is here in case you hear of anything similar. In trying to explain #2 above, I wonder if I perhaps had 2 tabs open and made one change in the 1st tab and the second in another. If I see anything more, I'll reopen the bug.
I am re-opening this bug, as I am seeing similar behavior to what is described above. I add a group of components and hit save. Status comes back and says the components were added, but a group of others was removed (and I had not made that change).
Just tested this on partner using Firefox and it seems to work properly.
Might have to get QE to bash on it to see if they can break it. qgong do you have time to do some testing on this?
(In reply to Whitney Chadwick from comment #2)
> I am re-opening this bug, as I am seeing similar behavior to what is
> described above. I add a group of components and hit save. Status comes
> back and says the components were added, but a group of others was removed
> (and I had not made that change).
I can' duplicate this issue too. I think you had 2 tabs open and made one change in the 1st tab and mad another other change in the 2nd tab. And you clicked the 'submit change' in the Mid-air collision page of the 2nd tab, so this will cause 1st tab's changes to be overwritten.
Feel free to contact with me when you are not agree with above idea.
I am confirming that I did not have 2 windows open. It has happened multiple times.
Raising priority, I'm seeing this again, this operation is not reliable.
I did NOT have 2 windows or tabs open, I was doing multiple ACL updates for RHEL 6.9 in the same tab and encountered the issue. Here is the output from the 2ND addition I tried:
Removed: java-1.8.0-openjdk, hwdata, gdb, device-mapper-multipath
Added: virt-manager, rsyslog7, virt-what, spice-vdagent, setroubleshoot
The "removed" list is the list of components I added in the previous update (about 30 seconds earlier).
Hi Rony, did you test this on a web server with a clustered DB? It's a shot in the dark, but it might be worth testing this on the QE cluster.
(In reply to Jeff Fearn from comment #7)
> Hi Rony, did you test this on a web server with a clustered DB? It's a shot
> in the dark, but it might be worth testing this on the QE cluster.
It isn't easy for me to test this on the qe cluster server,
connect to the same pgpool virtual host, So
1. I just access 05 or 06 and do the test, they works well
2. But when I access 05 and 06 in 2 tabs, and do changes in one tab(05), then do changes in another tab(06), there always prompt me the conflict exist.
Please note that we don't have proxy balance server that could let me access both 05 and 06 randomly.
The next time this happens can someone forward me the related emails? I'm hoping there is something in them that either gives me a hint of the problem or a hint of what narrow down a search through the logs.
(In reply to Tom Lavigne from comment #6)
> The "removed" list is the list of components I added in the previous update
> (about 30 seconds earlier).
Hmm this is interesting; due to the way Bugzilla processes input, if something is missing it's considered to have been deleted.
I wonder if there is some caching or something happening so that the browser doesn't update the "current approved components" list it's displaying, so the second commit looks like a delete because the components added in the first commit aren't in the list.
I can get this to happen inconsistently.
On Bugzilla 5 I get the mid air collision page:
Someone else has made changes at the same time you were trying to. The changes made were:
Who When What Removed Added
email@example.com 2016-12-08 05:08:46 UTC approved Cython, ConsoleKit, 389-ds-base
Run at 05:09:12 UTC.
hmm and if you just add things and pause then repeat eventually you get:
Who When What Removed Added
firstname.lastname@example.org 2016-12-08 05:14:08 UTC capacity DeviceKit-power
email@example.com 2016-12-08 05:14:37 UTC capacity DeviceKit-power
firstname.lastname@example.org 2016-12-08 05:15:30 UTC approved DeviceKit-power
email@example.com 2016-12-08 05:15:54 UTC approved Documentation
firstname.lastname@example.org 2016-12-08 05:16:18 UTC approved Driver Update Disk
I think this is due to the database servers having different times and the "select now()" being read can get sent to a server with an older time.
Hi Tom, Ops has checked the clocks on the databases and they are all in sync ATM, can you still reproduce this?
I can't see any other way this could be triggered, but if it's still happening for you I will beat on the code some more and see if anything else pops up. I can't reproduce on the devel cluster once the clocks are synced.
The QE cluster times are also in sync.
I have not seen the problem recently (probably since my last report), but I'm currently in the part of the release cycle where I only do occasional ACL updates, with minor changes in each. If I see it again, I'll report it.
Deferring until we can replicate this.