Rationale: if an module gets a resource that proves to be bad (e.g. a connection that's gone away), it shouldn't be returned to the pool to be given out again. We should invalidate it.
I'm proposing the following patch, though I'm not sure whether or not we should free_container in the event of destroy_resource returning an error(?)
--- apr-util/include/apr_reslist.h.old 2004-01-02 02:46:38.000000000 +0000 +++ apr-util/include/apr_reslist.h.new 2004-01-02 02:50:43.000000000 +0000 @@ -150,6 +150,15 @@ APU_DECLARE(apr_status_t) apr_reslist_release(apr_reslist_t *reslist, void *resource);
+/** + * Invalidate a resource in the pool - e.g. a database connection + * that returns a "lost connection" error and can't be restored. + * Use this instead of apr_reslist_release if the resource is bad. + */ +APU_DECLARE(apr_status_t) apr_reslist_invalidate(apr_reslist_t *reslist, + void *resource); + + #ifdef __cplusplus } #endif
--- apr-util/misc/apr_reslist.c.old 2004-01-02 02:44:05.000000000 +0000 +++ apr-util/misc/apr_reslist.c.new 2004-01-02 02:51:27.000000000 +0000 @@ -396,5 +396,17 @@
return reslist_maint(reslist); } +APU_DECLARE(apr_status_t) apr_reslist_invalidate(apr_reslist_t *reslist, + apr_res_t *res)
see comments in PR 26050 about discrepancy in prototypes that seems to show an underlying issue; apps don't see apr_res_t, so some rework and testing is needed
