Stas Bekman <[EMAIL PROTECTED]> writes:
I can't figure out whether PL_thr_key is a static global that survives perl_destruct and perl_free?
It is. because you find it declared in perlvars.h under the following heading: ---------------------------------------------------------->%------------- /****************/ /* Truly global */ /****************/
/* Don't forget to re-run embed.pl to propagate changes! */
/* This file describes the "global" variables used by perl * This used to be in perl.h directly but we want to abstract out into * distinct files which are per-thread, per-interpreter or really global, * and how they're initialized. * * The 'G' prefix is only needed for vars that need appropriate #defines * generated in embed*.h. Such symbols are also used to generate * the appropriate export list for win32. */
/* global state */ PERLVAR(Gcurinterp, PerlInterpreter *) /* currently running interpreter * (initial parent interpreter under * useithreads) */ #if defined(USE_5005THREADS) || defined(USE_ITHREADS) PERLVAR(Gthr_key, perl_key) /* key to retrieve per-thread struct */ #endif
[...]
Aha, I was grepping for the macro! Thanks.
If it does than this should work:
Index: src/modules/perl/mod_perl.c =================================================================== --- src/modules/perl/mod_perl.c (revision 124805) +++ src/modules/perl/mod_perl.c (working copy) @@ -573,6 +573,19 @@ MP_threads_started = 0; MP_post_post_config_phase = 0;
+ /* with USE_ITHREADS perl leaks pthread_key_t on every + * perl_destruct, which becomes a problem restarts: if the OS + * limit is 1024, 1024 restarts later things will start + * crashing */ + /* XXX: this is a workaround for the bug in perl, once and if it's + * fixed we need to disable it for the versions that have it + * fixed */ +#ifdef ALLOC_THREAD_KEY + if (PL_thr_key) { + FREE_THREAD_KEY; + } +#endif +
This would only work if the perl code is never re-entered after this. Is this really the case?
That's really the case
The '#ifdef ALLOC_THREAD_KEY' is wrong AFAICT; this macro will always be defined. Testing PL_thr_key as a boolean is outside of the spec of what pthreads define. PL_curinterp will be set if the key has initialized. That's a better test.
Does things not work if you just drop in the "perl_fini" destructor I provided in mod_perl.c?
It works too. The latest version is:
Index: src/modules/perl/mod_perl.c
===================================================================
--- src/modules/perl/mod_perl.c (revision 124805)
+++ src/modules/perl/mod_perl.c (working copy)
@@ -573,6 +573,17 @@
MP_threads_started = 0;
MP_post_post_config_phase = 0;+ /* with USE_ITHREADS perl leaks pthread_key_t on every
+ * perl_destruct, which becomes a problem restarts: if the OS
+ * limit is 1024, 1024 restarts later things will start
+ * crashing */
+ /* XXX: this is a workaround for the bug in perl, once and if it's
+ * fixed we need to disable it for the versions that have it
+ * fixed */
+ if (PL_curinterp) {
+ FREE_THREAD_KEY;
+ }
+
MP_TRACE_i(MP_FUNC, "mod_perl sys term\n");modperl_env_unload();
Thanks Gisle.
so the only remaining issue is how to fix that in perl-core so it works in any embed perl program.
-- __________________________________________________________________ Stas Bekman JAm_pH ------> Just Another mod_perl Hacker http://stason.org/ mod_perl Guide ---> http://perl.apache.org mailto:[EMAIL PROTECTED] http://use.perl.org http://apacheweek.com http://modperlbook.org http://apache.org http://ticketmaster.com
--------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
