Details
-
Bug
-
Status: Closed
-
Major
-
Resolution: Fixed
-
5.3.0, 6.0.0
-
RHEL 6.x, CentOS 6.x (on x86_64)
Description
I recently upgraded from trafficserver 4.x to 5.3.0, and one thing that I noticed was that certain traffic_line commands started printing "requested command failed". For example:
# trafficserver start Starting Apache Traffic Server: [ OK ] # traffic_line --status Proxy -- on # traffic_line --shutdown error: the requested command failed: [11] Invalid parameters passed into function call. # traffic_line --startup error: the requested command failed: [11] Invalid parameters passed into function call.
Interestingly, even if the --shutdown command gives such an error message, the diags.log file still shows that the server has shut down:
[May 28 22:03:03.539] Server {0x2abe4251d700} NOTE: [ProcessManager::processEventQueue] Shutdown msg received, exiting
and the --status command also finds the same:
# traffic_line --status Proxy -- off
Similar for the --startup command: though it gives the same error message, the command seems to have actually worked.
I did some bisecting to find where this was introduced, and ended up at commit 355c165 (TS-3033: fix PROXY_STATE_GET). Before this particular commit, both traffic_line --shutdown and traffic_line --startup work without printing any error; after the commit, they both start printing "requested command failed".
If I look at all the work done in TS-3033, and the particular change in commit 355c165, I would guess that some parts of the code are disagreeing about the number of parameters for the PROXY_STATE_GET message.
Alternatively, there may be some sort of issue in the management API and/or marshalling code, where a PROXY_STATE_GET message with two parameters gets "cut off" in some way, possibly because the server is just shutting down at that point?
Attachments
Attachments
Issue Links
- relates to
-
TS-3033 Reimplement management protocol
- Closed