Intermittent crash with logging to DB enabled

[paulh]

The attached BT shows a crash in mytharchivehelper but I believe it affects all the command line tools that start up and exit quickly.

[paulh]

mytharchivehelper crash running 'mytharchivehelper -p --output=test.txt'

[danielk]

It looks like gContext is never deleted in most returns from mytharchivehelper's main function.

The MythContext destructor doesn't just shut down the logging thread. There are at least a half dozen other ways a program that constructs MythContext and never deletes it can segfault just before exit.

[danielk]

Paul, I don't know how one runs mytharchivehelper, but the attached patch is a pretty boilerplate fix for this type of object lifetime problem.

[Github]

Add CleanupGuard to mytharchivehelper

Refs #10085

Signed-off-by: Gavin Hurlbut <ghurlbut@…>

Branch: master Changeset: cfeca636694219d4dd7ef021be5e3f12d57c5693

[paulh]

Crash after cleanup guard added

[paulh]

Adding the cleanup guard has made no difference still getting the same crash. New bt added.

[beirdo]

OK. Could you bring up gdb with that coredump again, and do:

thread 1
up

and continue to go up until you are at #5 which is the logqmsg at logging.cpp:651 or so.

Then

print *info
print query

Also, what is your locale set to?

[paulh]

(gdb) thread 1
[Switching to thread 1 (Thread 28537)]#0  0x0000000000000000 in ?? ()
(gdb) up
#1  0x00007f5b24ee6327 in QTextCodec::fromUnicode(QString const&) const () from /usr/lib64/qt4/libQtCore.so.4
(gdb) up
#2  0x00007f5b19de58fc in ?? () from /usr/lib64/qt4/plugins/sqldrivers/libqsqlmysql.so
(gdb) up
#3  0x00007f5b266d08f6 in QSqlQuery::exec() () from /usr/lib64/qt4/libQtSql.so.4
(gdb) up
#4  0x00007f5b295a47f1 in MSqlQuery::exec (this=0x721c60) at mythdbcon.cpp:651
651         bool result = QSqlQuery::exec();
(gdb) up
#5  0x00007f5b29648041 in DatabaseLogger::logqmsg (this=0x6a8240, query=..., item=0x6b51e0) at logging.cpp:462
462         if (!query.exec())
(gdb) print *info
No symbol "info" in current context.
(gdb) print query
$1 = (MSqlQuery &) @0x721c60: {<QSqlQuery> = {d = 0x71d3c0}, m_db = 0x6c0d70, m_isConnected = true, m_returnConnection = true, 
  m_last_prepared_query = {d = 0x6b3620}}
(gdb) print item.message 
$2 = "Added database logging to table logging", '\000' <repeats 1889 times>
(gdb)

$ locale
LANG=en_GB.UTF-8
LC_CTYPE="en_GB.UTF-8"
LC_NUMERIC="en_GB.UTF-8"
LC_TIME="en_GB.UTF-8"
LC_COLLATE=C
LC_MONETARY="en_GB.UTF-8"
LC_MESSAGES="en_GB.UTF-8"
LC_PAPER="en_GB.UTF-8"
LC_NAME="en_GB.UTF-8"
LC_ADDRESS="en_GB.UTF-8"
LC_TELEPHONE="en_GB.UTF-8"
LC_MEASUREMENT="en_GB.UTF-8"
LC_IDENTIFICATION="en_GB.UTF-8"
LC_ALL=

[beirdo]

Argh, I did mean item, not info. My bad. You got us what I was looking for anyways. Thanks. Hmm. Could you attach the debug logs (which hopefully you had running) for the crashed process? It seems to be a unicode vs utf-8 issue, which I believe there has been some extra diagnostic logging added fairly recently.

[paulh]

Added a log. Just to remind you it doesn't crash every time only about 1 out of 6 attempts.

[beirdo]

I can't recreate this issue with current master. Is this fixed? If not, do you have another commandline that exhibits the issue so I can retest? In 20 runs of the command given (with --output replaced with --outfile), not a single crash.

[beirdo]

5 weeks with no confirmation... Closing this.