GIT Log for 2009-11-30 00:33 GMT
aland at deployingradius.com
Mon Nov 30 01:33:02 CET 2009
Author: Alan T. DeKok <aland at freeradius.org>
Date: Sun Nov 29 16:07:23 2009 +0100
Clean up state machine.
This error happens when "max_request_time" is set VERY low.
i.e. lower than "response_window". (12s versus 30s).
The current logic for enforcing the various timers is pretty bad. There
is one timer per request, and it bounces around between the different
requirements. At the time it was written, it seemed simpler than trying
to manage 3-4 simultaneous timers per request.
When the request is proxied, the timer being applied is for
"response_window". BUT by the time that expires, the "max_request_time"
has expired. The code *does* notice that it has expired. BUT it doesn't
notice that there's no child thread processing the request. So it waits
for the child thread to exit... forever.
At some point, a timer overflows, and it dies.
There are a few changes to make:
1) check for "no child" in this situation, and clean up the request rather
than waiting forever.
2) cap the timer to 5 minutes (this can still happen, for example, when a
bad DB locks a thread for hours at a time).
3) don't overflow when adding timer values.
src/main/event.c | 64 ++++++++++++++++++++++++-----------------------------
1 files changed, 29 insertions(+), 35 deletions(-)
More information about the Freeradius-Devel