Hello,
The attached patch prevents TcpAcceptor job leaks on reconfiguration
by monitoring and reacting to the listening socket closure.
This monitoring does not happen automatically via COMM_ERR_CLOSING
because TcpAcceptor does not use comm_read and comm_write APIs. It has
to use the raw SetSelect() API instead.
The patch is against a v3.3-based branch but seems to apply to trunk
with enough fuzz. I can massage the patch before commit if approved. I
have not tested with trunk, but the TcpAcceptor code appears to be the
same in this area -- it has no explicit close handlers registered. If
you do not know of any implicit Comm fixes, this bug is probably still
present.
HTH,
Alex.
This archive was generated by hypermail 2.2.0 : Thu Feb 13 2014 - 12:00:12 MST