2017-06-23 16:20 GMT-05:00 Mark Thomas <ma...@apache.org>: > On 23/06/17 17:52, l...@kreuser.name wrote: > >> Am 23.06.2017 um 17:32 schrieb Mark Thomas <ma...@apache.org>: > > <snip/> > > >>> However I still see a timing problem, as it is not always happening. > Maybe 5 out of 7 are NULL. If I send more requests immediately after each > other NULL is more frequent. > >> > >> Thanks for the confirmation. The important thing is that the crashes > >> have stopped. Note I'm fairly certain this is issue only affects NIO2 > >> with OpenSSL for TLS. > >> > >> I'm currently looking at a way to pre-populate the TLS info if the > >> AccessLogValve requires this to avoid the error messages and the NULL > >> cipher suite entries. > >> > >> Mark > >> > > > > I agree with that - no more crashes is key! > > > > Thanks. Have a great weekend. > > The ISEs are fixed. Actually the test has been removed. It is possible > to see null session IDs with OpenSSL if session tickets are enabled. > > I've also added some code to the AccessLogValve that pre-populates the > TLS attributes if the access log is configured to log them. > > So, as far as I can tell, everything should now be fixed. > > The complete set of fixes will be in 9.0.0.M23 onwards and 8.5.17 onwards. > > Ok, so I'll add my patch for APR to fix 60461 then ?
IMO, until the connection close waits for all streams to complete, which may not be the best idea, the sync is going to be needed for all connectors. Rémy