[Sbcl-help] SBCL + CLSQL (postgres) drops into ldb

[Simon Koch]

I am currently running into some major problems using a combination of
sbcl 1.2.0 (Linux x64 Ubuntu) and clsql-20130420 (using postgres and
installed via quicklisp).

After running my program for a while (between 20 and 60 minutes) it
crashes with the message:
Signal 13 masked
fatal error encountered in SBCL pid 57022(tid 140736637269760):
some deferrable signals blocked, some unblocked

Error opening /dev/tty: No such device or address
Welcome to LDB, a low-level debugger for the Lisp runtime environment.

A backtrace reveils:
ldb> backtrace
Backtrace:
0: Foreign function lose, fp = 0x7fffcd455810, ra = 0x41279e

1: Foreign function (null), fp = 0x7fffcd4558e0, ra =
0x412f86
2: Foreign function check_interrupt_context_or_lose, fp = 0x7fffcd455920,
ra = 0x413db2
3: Foreign function maybe_defer_handler, fp = 0x7fffcd455970, ra = 0x413fb4
4: Foreign function (null), fp = 0x7fffcd4559b0, ra = 0x4160fd
5: Foreign function (null), fp = 0x7fffcd455f30, ra = 0x7ffff79c8cb0
6: Foreign function (null), fp = 0x7fffcd456120, ra = 0x7fffdecab36f
7: Foreign function (null), fp = 0x7fffcd456140, ra = 0x7fffdeca2547
8: Foreign function PQgetResult, fp = 0x7fffcd456180, ra = 0x7fffdeca05a5
9: Foreign function (null), fp = 0x7fffcd4561b0, ra = 0x7fffdeca0848
10: (SB-PCL::FAST-METHOD CLSQL-SYS::DATABASE-EXECUTE-COMMAND
(COMMON-LISP::T QL-DATABASE))
11: (SB-PCL::EMF CLSQL-SYS::DATABASE-EXECUTE-COMMAND)
12: (SB-PCL::FAST-METHOD CLSQL-SYS::EXECUTE-COMMAND (COMMON-LISP::STRING))
13: (SB-PCL::FAST-METHOD CLSQL-SYS::EXECUTE-COMMAND
(CLSQL-SYS::%SQL-EXPRESSION))
14: (COMMON-LISP::FLET WITHOUT-INTERRUPTS-BODY-645 KEYWORD::IN
SB-THREAD::CALL-WITH-MUTEX)
15: SB-THREAD::CALL-WITH-MUTEX

Sadly the error is not replicable, just executing the same query does not
lead to the exception and is actually done several dozend times
beforehand. It also does not have to be the same query in the next run.
The only connection between the queries/contexts of the crashes I can make
out is that they are heavily used (multiple times per minute).

This problem seems to exist for quite some time:
http://comments.gmane.org/gmane.lisp.lib.postmodern.devel/156

But I was not able to find any solutions.

As I cannot reduce this problem to a proof of concept I am not sure how to
proceed and I'd be glad for any suggestions on how to isolate the code
that triggers that behavior and work around it as it currently kills my
application.
I first encountered the problem using sbcl-1.1.12 and then upgraded to 1.2.0.

Thanks,

Simon K.


------------------------------------------------------------------------------
HPCC Systems Open Source Big Data Platform from LexisNexis Risk Solutions
Find What Matters Most in Your Big Data with HPCC Systems
Open Source. Fast. Scalable. Simple. Ideal for Dirty Data.
Leverages Graph Analysis for Fast Processing & Easy Data Exploration
http://p.sf.net/sfu/hpccsystems
_______________________________________________
Sbcl-help mailing list
Sbcl...@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/sbcl-help

[Waldek Hebisch]

Simon Koch wrote:
>
> I am currently running into some major problems using a combination of
> sbcl 1.2.0 (Linux x64 Ubuntu) and clsql-20130420 (using postgres and
> installed via quicklisp).
>
> After running my program for a while (between 20 and 60 minutes) it
> crashes with the message:
> Signal 13 masked
> fatal error encountered in SBCL pid 57022(tid 140736637269760):
> some deferrable signals blocked, some unblocked

AFAICS sbcl insists that signal mask is set in a special way. If
external C code uses signals and sets signal mask, then this
condition may be violated. It would be good if sombody knowing
sbcl internals explained which signals are reserved for sbcl
use and what to do when C code wants to use signals. Similarely
for postgres interface library (I hope that postgres documents
its signal use, but I did not check).

--
Waldek Hebisch
heb...@math.uni.wroc.pl