Shared publicly  - 
Almost threw my laptop out the window last night.

Started attempting to write my monthly status report. Want to Cc it to the 'all department' mailing list, so start typing the address, wait for auto-completion (it changes frequently so I don't remember what it is this week). Gr, Evolution's address book and calendar seem to stop working every time I drop off the VPN and don't start working again when I rejoin. Probably related to in some way (a similar bug, if not a symptom of the same bug).

OK, restart evolution-addressbook-factory. Hm, it needs to download a new version of the addressbook. Oh, and for some reason the incremental update doesn't work (that one probably my fault) and it downloads a completely new copy. And then copies its 121MiB of data from the downloaded file into a sqlite database.

Now, Linux's performance under any kind of I/O load is really sucky, but this was even suckier than normal. During this period it took me three minutes just to focus on the xchat window and type a simple response to a question. Normally it's only a minute or two — and I'm not sure how much of that to blame on gnome-shell. Yes, I do have /proc/sys/vm/dirty_ratio set to 1, not the default 20.

I run 'iotop', and half the time it seems to be telling me there is no I/O. That's blatantly nonsense. Eventually I find that abrt is running — taking a core dump of Evolution, which has crashed. So it's writing the 4½GiB core dump to disk. Yes, four and a half fucking gigabytes. I have no idea what it's using it all for. I do periodically run it under Valgrind, and have experimented with some GObject trackers too, but there is no smoking gun; it just grows and grows and grows (as does Thunderbird, I'm told).

I ran gnome-abrt and started trying to file the report. But by the time it'd finished trying to process it, the fucking thing had been deleted! It seems that Yet Another Gnome Shell crash (qv) had caused the evolution dump to be removed while it was still being processed:

Nov 18 00:02:00 abrt-hook-ccpp[19897]: Saved core dump of pid 13172 (/usr/bin/gnome-shell) to /var/tmp/abrt/ccpp-2013-11-18-00:01:17-13172 (384856064 bytes)
Nov 18 00:02:00 abrt-hook-ccpp[19897]: /var/tmp/abrt is 1642394496 bytes (more than 1279MiB), deleting 'ccpp-2013-11-17-22:54:19-2299'

Thankfully, I had also loaded gdb on the coredump manually, and still had it open so I was able to keep a copy.

But even after all this crap had finished, the machine still hadn't recovered from the I/O load, so I gave up and went to bed. Twenty minutes later I came downstairs for Bonjela and Calpol and prodded it, and it still didn't come back. The backlight came on, but it just sat there at a black (but backlit) screen the whole time I finding and measuring out a syringe of medicine. Perhaps that was partly the normal shell crash on waking from idle, but probably not.

I no longer give Linux machines to my family to use. I still try to support the ones I have given out in the past, but it is increasingly painful — just as painful as it is trying to use Linux myself on the desktop.

This sure as hell isn't the year of the Linux desktop. It might just be the year of the MacOS desktop, for me.
GNOME Bugzilla – Bug 708175. "You must be working online to complete this operation". Last modified: 2013-11-05 09:54:25 UTC. Home; | New; | Browse; | Search; |. | Reports; | Requests; | Help; | New Account; | Log In. Remember [x]. | Forgot Password. Login: [x]. First Last Prev Next No search ...
Dan Carpenter's profile photoMauro Andreolini's profile photoAndreas Tunek's profile photoDavid Woodhouse's profile photo
I'm wondering what could be done about this. I found Ubuntu's desktop much nicer to use and less resource-intensive; but am reluctant to import the packages of Unity 7 (they exist though) to Fedora due to uncertain future upstream. XFCE spin might be a good idea too?

Abrt's not in default Ubuntu installation either; probably it is a better desktop out of box?

When it comes to Evolution, I've found out it deals fine with smaller volume of mail, and sucks heavily (memory consumption, generally slow) for big loads. I'm wondering how relevant is it these days compared to web gmail or outlook live or whatever.

By the way, I've lowered my dirty_bytes watermark to a lower value to cope with i/o freezes, and everything seems to work nice apart from that it makes yum (surprisingly nothing else I use seems to cause that much disk traffic) unusable.
Core'ing big processes does seem to really screw things up, even without things like abrt.
Right. I'm wondering if would be too difficult to dump the core after the process has terminated in case the core collector is a pipe, (parent's wait() would finish and file descriptors and other resources but vmas torn down), so that the crahsed process' window does not sit stuck on the desktop. Also the limit on core size could be enforced by kernel, so that poor abrt does not even bother trying to dump something insanely big.
I was OK with the crashed process' window sitting stuck on the desktop. The problem was that I couldn't use the damn desktop at all anyway, because the whole damn machine sits there for minutes at a time failing even to focus a new window, when there's high I/O load. And then after writing out a 679505920-byte core file (seriously, Evolution, wtf was all that), the poxy abrt system deleted it again before I could do anything useful with it!
+David Woodhouse I seriously suggest that you install something else instead of Evolution. I tried to use it during GUADEC, and couldn't - if the VPN drops while it sends or receives e-mail, it doesn't respond even to closing the window. Try Thunderbird, despite all that bad publicity that you quoted - at least it does understand that cancel means cancel.

As for the suggestion to use XFCE - maybe it is a good idea, maybe not. The plus is that it doesn't really force any particular applications as a strongly suggested default. You use whatever e-mail client and whatever browser works for you, while still being sure that no "desktop integration" is being lost (because there was none to begin with). The minus is that the built-in configuration tools (even for such basic things as keyboard and monitors) suck badly, and you'll resort to command-line tools more often than you would in the ideal world. And there is no theme that is consistent between gtk2 and gtk3 >= 3.8 that is not Adwaita.
And there I was this morning, pondering whether to switch to Evolution because of Thunderbird sucking more than used to :(

About those big core dumps, I guess it's mapped files such as mailbox indexes?
It seems to be sucking a lot more with 3.10 for offline use, it at least use to detect VPN up/down and reconnect without restart.
I abandoned desktop Linux for similar reasons around 2004, but gave it another chance a month ago. Replaced the dread Windows 8 with Mint on two of my regular machines, a Thinkpad X1 and the Office desktop.

I was surprised how well it works out of the box. Even the font rendering looks reasonably good.
+Alexander Patrakov Do you have a bug number for that?
+Tomeu Vizoso Although I had a bad day yesterday, Evolution isn't actually that bad. And I am a demanding user. It has good IMAP support, and supports Exchange Web Services for my company email/calendar fairly nicely too with the odd caveat about offline/vpn stuff occasionally. Its memory usage is high but I'm told that's true of Thunderbird too. I'm generally fairly happy with it.
+Tom Kistner I actually found it a whole lot less frustrating in 2004. These days I feel we have lost the technical edge and the thirst for excellence. It's all thrown together hurriedly without enough attention to detail. And I'm as much to blame as anyone else, I'm sure.
I've been feeding people Ubuntu with Unity. Strange as it may sound the non technical recipients seem to adore it. For anything else there is Cinnamon.
+David Woodhouse no bug report yet. If I find enough time to upgrade the system on the partition where I tested GNOME, I will do that. Otherwise (i.e. very likely, as I am no longer a GNOME user), no, as all bug reports about the previous version are rightfully ignored.
I've run into these kinds of behavior a lot, leading to lots of frustration... usually when things get bloated enough to start "swapping" (well, paging to disk)... which seems a symptom of the same problem, performance under I/O load.  More RAM makes it worse when you hit the conditions: the leaky programs eat up that much more space before they page or have to dump a monster core, or have something processed by abrt, or...  Are we working on this problem? My feeling is 10 yrs ago, less bloat, less memory, so shorter storms when the disk hasta get really busy in case of hiccups. Normal-state stuff may be better, but edge cases have gotten worse.
The problem is the behaviour of the kernel when dumping core. There seems to be no way to throttle anything, and I'm not sure whether it's not trying to fill the pagecache with that core. It's not a new problem...
If you're doing a huge write to disk then linux totally does suck...  I have this problem every day because I create a 4gb sqlite database.

Part of the problem is that by the time the data hits the disk it doesn't know which process sent the data so it can't schedule fairly and switch between processes which need to write.

Part of the problem is that gnome requires tons of stupid writes to disk before the screen will refresh.
One is tempted to be catty and suggest that a sentence which includes both 4gb and sqlite can only exist if it also includes invective :)
sqlite is fine if you use it like i do with just one reader and no writers.  It would probably work fine with multiple readers as well...

It's only if you try reading and writing at the same time that it doesn't work.  ;)
I like the standard internet responce to a hard problem, "Use something completely different!". How would using XFCE or Unity ever help when this problem seems to be the interaction between Evolution and the kernel Linux?

And to Steve, how would you fix bugs if you stop development?
If you are happy with your old distros, good for you. But I did not see anything in your reply regarding how bugs should get fixed. Bugs don't get fixed by waiting, the get fixed by someone working on them.
I remember my sound card working properly with a newer Linux (Linux has a 3-4 month release cycle right?).
For my corporate users, the SSL certificate handling in Fedora 19 is a must-have. Installing the company's certs once and expecting them to Just Work, systemwide - not only for some apps, or some crypto libraries. Up to date and well-maintained (not just 'packaged') versions of other things like Evolution-EWS and pidgin-sipe are also extremely useful. The GSS-NTLMSSP module will land I Fedora first and be properly integrated there with a holistic view of the distro too, I'm sure...
Add a comment...