Demo Discussion
Forum Config Examples Contributions Vulnerabilities
  Discussion forum about ELOG, Page 100 of 806  Not logged in ELOG logo
ID Date Icon Authordown Author Email Category OS ELOG Version Subject
  66456   Wed Jul 22 12:15:56 2009 Reply T. Ribbrockemgaron+elog@ribbrock.orgBug reportLinux2.7.6r2233Re: Crashes when editing entries

T. Ribbrock wrote:

For some odd reasons, we are experiencing frequent crashes of elogd over the past few days. It has been working fine so far, but more or less out of the blue it became rather unreliable. The current configuration is installed on two servers, one running 2.7.5.-r2174 on ClarkConnect 4 and one running 2.7.6-r2233 on Debian 4.0 - both show the same problem. Each of them has an "active" group with four logbooks and an "archive" group with three logbooks. In the "active" group, there are two logbooks that share the same index (using Subdir=...) and it looks like the crashes occur most of the time in these, though that's just a hunch so far. Also, most of the crashes seem to happen when submitting an entry that has been edited. Actually, submitting a modified entry has always been strange in our logbooks: When we hit submit, we get a pop-up window asking "Submit modified entry?". When choosing "OK", the entry that has been edited is duplicated. When choosing "Cancel", it is submitted correctly.

I've been running elogd like this (to get more info)

elogd -v > elog-2233-2.log 2>&1

The last entry I get in the log when elogd crashes is:

  Same index as logbook Machine Log
elogd: src/elogd.c:727: xfree: Assertion `*((unsigned int *) (temp - 4)) == 0xdeadc0de' failed.
Received unknown cookie "wikidb_mw__session"
Received unknown cookie "wikidb_mw__session"

I did actually make a few changes to the configuration before we noticed the crashes: I added one extra attribute and a few more conditionals.

 

Any additional information you need: Just let me know.

Regards,

Thomas

 Forgot to mention: I've also seen error messages like this upon a crash:

*** glibc detected *** corrupted double-linked list: 0x0911bbc0 ***

Regards,

Thomas

  66458   Wed Jul 22 15:35:57 2009 Reply T. Ribbrockemgaron+elog@ribbrock.orgBug reportLinux2.7.6r2233Re: Crashes when editing entries

Stefan Ritt wrote:

well, I need to reproduce your problem in order to fix it. The failed assertation you get is due to some internal writing beyond array boundaries, but I have no clue which part of the code makes this. It might be related to the fact that you use the same index (via Subdir=...) for two logbooks. In this scenario, you are only allowed to modify/add entries to one logbook, not the other. The other one may only be used for reading. And even then it's not guaranteed that new entries show up in the second logbook immediately, you might have to restart the server in order to re-index the logbooks. Internally, the daemon does not know that two logbooks are "the same" and one instance will not realize if the other instance modifies the data "below its feet". Can you try to give up the double logbooks and see if the problem goes away?

 Hm... I have implemented this set-up originally based on this: https://midas.psi.ch/elogs/Forum/66024. The "double logbook" is a machine log with a "software" (OS installations etc.) and a "hardware" (CPU, RAM, etc.) view. The "hardware" view has the "Subdir=" statement. Thinking about it, the "software" view is used most - I have several automatic scripts running which update the contents whenever a machine gets updated, re-installed and so on. The hardware part does not see much editing - until this week, when we decided to start an inventory... So, it's quite possible that we never noticed that this was iffy. For the rest of our goals, this set-up has worked fantastically - never noticed any problem with one view not updating, actually. Also, I do not remember any crashes with the other, single logbooks.

What I've done for now is to ask all team members to use only the software part (the one without the Subdir statement) to actually change content (the entry masks are the same in both versions) and use the hardware part just for viewing. I'll report back as soon as I get some feedback.

Nonetheless, given that this set-up has been a great help for us - if you ever get the chance to make this work (even) better, I'd be most grateful.

Regards,

Thomas

  66460   Wed Jul 22 16:52:13 2009 Reply T. Ribbrockemgaron+elog@ribbrock.orgBug reportLinux2.7.6r2233Re: Crashes when editing entries

Stefan Ritt wrote:

T. Ribbrock wrote:

Nonetheless, given that this set-up has been a great help for us - if you ever get the chance to make this work (even) better, I'd be most grateful.

Well, for that I have to reproduce the problem. So best would be if you strip it down to the bare minimum in order to reproduce this reliably. Then you zip everything and send it to me. Then tell me what I have to edit and submit in order to stimulate the crash. Once this is successful, I can fix it.

Thank you - I shall look into that, though it'll probably take a while to prepare it.

  66464   Mon Jul 27 10:20:14 2009 Entry T. Ribbrockemgaron+elog@ribbrock.orgBug reportLinux2.7.6Wrong error message if invalid attribute is used

I just ran into this little bug: I had defined a new logbook in my config file and suddenly got the message Attribute "Date" not allowed. While I did have several attributes starting with the word "Date" (e.g. "Date In Service", "Date Retired") I had no attribute "Date" in there. After some pondering and wildly commenting out lines, it finally dawned on me: I had used an attribute "ID" - which is also not allowed. However, it would be very helpful if the error message actually reflected that...

  66472   Tue Jul 28 16:30:46 2009 Entry T. Ribbrockemgaron+elog@ribbrock.orgBug reportLinux2.7.6r2233Odd problem (bug?) with certain attribute

I have the following simple test logbook:

; General
List display = Edit, Hostname, OS, Size
Entries per page = 150
Quick filter = OS
Date Format = %d/%m/%Y
Summary Lines = 0

; Attributes
Attributes = Hostname, OS, CPU,Size
Required Attributes = Hostname
Sort Attributes = Hostname

; Message part: log is text only, but elog is allowed
Default encoding = 0
Allowed encoding = 2

For some strange reasons, I'm having problems with the "Size" attribute, which I have added later. If I start adding/editing entries, at some point, the "Size" attribute will stay at "0" and will not accept any further changes. I've tried to pare down the config to maybe find which statement could be causing this, but to no avail. The only thing I can say is taht id doesn not seem to happen if the configuration consist of the sole line

Attributes = Hostname, OS, CPU,Size

I'm quite puzzled as to what is going on here (the original problems stems from a more complicated logbook) - I'm not even 100% sure whether this happens due to a bug or not.

 

Regards,

 

Thomas

  66474   Tue Jul 28 17:10:13 2009 Reply T. Ribbrockemgaron+elog@ribbrock.orgBug reportLinux2.7.6r2233Re: Odd problem (bug?) with certain attribute

Stefan Ritt wrote:

That's indeed a strange bug, and thanks to your detailed explanation I could easily reproduce it. The problem was that in the ELCode toolbar there is already a "SIZE" parameter, which gets submitted instead of your "size" attribute. Therefore whatever you submit as "size", gets replaced by zero (since the SIZE drop-down box usually sits at zero). So you can either go and change your "size" attribute into someting else like "Memory Size", or you upgrade to version 2244, where I fixed this problem.

 Very nice, thank you! Given that I also still have to test the fix you made for the crash problem with "shared" logbooks (I assume it's also present in 2244), I'll upgrade and report back (probably tomorrow).

  66482   Wed Jul 29 13:23:41 2009 Agree T. Ribbrockemgaron+elog@ribbrock.orgBug reportLinux2.7.6r2233Re: Odd problem (bug?) with certain attribute

T. Ribbrock wrote:

 Very nice, thank you! Given that I also still have to test the fix you made for the crash problem with "shared" logbooks (I assume it's also present in 2244), I'll upgrade and report back (probably tomorrow).

 2244 is now running, and indeed, the "Size" attribute works now. Thank you!

  66483   Wed Jul 29 14:48:34 2009 Agree T. Ribbrockemgaron+elog@ribbrock.orgBug reportLinux2.7.6r2233Re: Crashes when editing entries

By now, I've installed 2244 and ran some rudimentary tests. So far, I was not able to reproduce the crash anymore. Looking good!

ELOG V3.1.5-3fb85fa6