[Date Prev][Date Next]
[Chronological]
[Thread]
[Top]
ITS#5835 insights?
Does anyone have any thoughts on ITS#5835?
The server keeps hitting the assert in connection.c:
static void
connection_close( Connection *c )
{
ber_socket_t sd = AC_SOCKET_INVALID;
assert( connections != NULL );
assert( c != NULL );
/* ITS#4667 we may have gotten here twice */
if ( c->c_conn_state == SLAP_C_INVALID )
return;
assert( c->c_struct_state == SLAP_C_USED );
--> assert( c->c_conn_state == SLAP_C_CLOSING );
Which was called by connections_timeout_idle in connection.c:
if( difftime( c->c_activitytime+global_idletimeout, now) <
0 ) {
/* close it */
connection_closing( c, "idletimeout" );
--> connection_close( c );
i++;
With the modified debug logging Howard checked in, I see the following
steps taking place:
Dec 2 03:06:05 ldap slapd[8292]: connection_closing: readying conn=3049
sd=28 for close
so we know it made it into connection_closing.
Then we see:
Dec 2 03:06:05 ldap slapd[8292]: connection_resched: attempting closing
conn=3049 sd=28
which means it made it to connection_resched, which calls connection_close:
Debug( LDAP_DEBUG_CONNS, "connection_resched: "
"attempting closing conn=%lu sd=%d\n",
conn->c_connid, sd, 0 );
connection_close( conn );
Doesn't this directly conflict with line 242 above, which also calls
connection_close on the connection?
I also see this comment in the connection_closing code:
/* ITS#4667 this may allow another thread to drop
into
* connection_resched / connection_close before we
* finish, but that's OK.
*/
I'm honestly not sure this is really "OK". The only place I see
connection_resched called is from connection_operation, and I don't see
exactly how I'ved ended up in there in this code path.
--Quanah
--
Quanah Gibson-Mount
Principal Software Engineer
Zimbra, Inc
--------------------
Zimbra :: the leader in open source messaging and collaboration