Project

General

Profile

Bug #3051

4.0.5-17~g15506ea - segfault while tuning bouquest

Added by Dreamcat 4 about 6 years ago. Updated about 6 years ago.

Status:
New
Priority:
Normal
Assignee:
-
Category:
Crashes
Target version:
-
Start date:
2015-08-23
Due date:
% Done:

0%

Estimated time:
Found in version:
4.0.5-17~g15506ea
Affected Versions:

Description

Hi.
I got a segfault today while tuning. After clearing config, upgrading to the latest 'testing' release (of release/4.0 branch).

So I restarted it again with debugging image (same version but with the _dbg symbols). And it errored out almose immediately.

Backtrace here (also attached to this ticket):

https://gist.github.com/dreamcat4/051e10c4c3818dd69465

Full debug trace log here (9.6mb, slow to load), also attached here as .zip file):

https://gist.githubusercontent.com/dreamcat4/67b634852cfb9b618f40/raw/8e8d5ee6a1be563904fd777bf5dee63a581664fe/tvh.debug.log

Many thanks if you can help to solve it. Because I must rely on the testing branch since there was no `4.0.6` released. Some important fixes were added since v4.0.5.


Files

gdb-backtrace.log (21.9 KB) gdb-backtrace.log <----- Please see here first Dreamcat 4, 2015-08-23 12:25
terminal.log (9.51 MB) terminal.log 9.6mb Dreamcat 4, 2015-08-23 12:27
terminal.log.zip (621 KB) terminal.log.zip 621k Dreamcat 4, 2015-08-23 12:27

History

#1

Updated by Dreamcat 4 about 6 years ago

I have now rolled back to 4.0.5-13~g7ca4286. And continued the same scan from where it left off. It's been going for about 10 minutes.

Waiting to see if 4.0.5 +13 ver also segfault. If not segfault, i don't report anything new or update this ticket after 1-2 day. Then it may be a new issue in the most recent commit(s). Please investigate if you can.

#2

Updated by Jaroslav Kysela about 6 years ago

Could you run valgrind on 4.0.5-17 ? There are no changes which should invoke a segfault in diff 13-17..

#3

Updated by Dreamcat 4 about 6 years ago

Jaroslav Kysela wrote:

Could you run valgrind on 4.0.5-17 ? There are no changes which should invoke a segfault in diff 13-17..

Ok. Never used valgrind before. I tried to run valgrind but not sure if it worked properly. Because then there was no output from tvheadend-dbg. And it exited very fast.

https://gist.github.com/dreamcat4/af947209b3b031cf9008

Would like to try again / better. Just not very good at this.

#4

Updated by Dreamcat 4 about 6 years ago

Dreamcat 4 wrote:

And it exited very fast.

Ah. Maybe i need to remove `--fork` from my $tvheadend_args.

#5

Updated by Dreamcat 4 about 6 years ago

Dreamcat 4 wrote:

Maybe i need to remove `--fork` from my $tvheadend_args.

https://gist.github.com/dreamcat4/730507fba35919534c72

Another fast exit during startup. Again I apologise. Don't know if that was what you were looking for.

#6

Updated by Dreamcat 4 about 6 years ago

Ok. I eventually got valgrind to work while tvheadend continues to run. But then could no longer reproduce any segfault crash.

I am suspicious of my config. Since it turned out I had 3 bouquets enabled all together. the 'SD', 'HD' and 'G2' versions of same region. I disabled the 2 not-needed bouquets in subsequent testing. So it was only tuning with 1 bouquet in the subsequent sessions. Then no segfault has occurred since that time.

#7

Updated by Jaroslav Kysela about 6 years ago

It looks like a memory corruption somewhere, but it's very difficult to trace it..

#8

Updated by Dreamcat 4 about 6 years ago

Ok perexg. Went back to ~17 with 1 bouquest enabled in config. It's been scanning fine. No crashes.

So I'm going to assume / attribute the issue to be because of scanning with multiple bouquets enabled (in my case it was 3 of them). The 3 bouquets were all very similar to each other, with 90% + same channels. So maybe there was some contention going on. For exaxmple perhaps if 2 different bouquet were trying to update the same channel simultaneously. It's just a speculation.

Anyway. I'm glad we have eliminate ~13 to ~17 changeset being anything to do with it. And for myself - it seems I only need 1 bouquet enabled. As terrestrial (DVB-T) my other input source don't use bouquets at all.

Many thanks for your help perexg. Please feel free to close this issue / as you wish.

#9

Updated by Dreamcat 4 about 6 years ago

I might also be so bold to ask you to please tag 4.0.6 from the release/4.0 branch. If there are not any other issues / pending changes. Because it includes an important fix regarding the seperate user DVR profiles / DVR access. Which came after the v4.0.5.

Also available in: Atom PDF