Bugzilla – Full Text Bug Listing
|Summary:||inefficient call of ps/netstat in verify_sessions handler|
|Product:||ThinLinc||Reporter:||Samuel Mannehed <firstname.lastname@example.org>|
|Component:||VSM Agent||Assignee:||Henrik Andersson <email@example.com>|
|Status:||CLOSED FIXED||QA Contact:||Bugzilla mail exporter <firstname.lastname@example.org>|
|Priority:||P2||CC:||email@example.com, firstname.lastname@example.org, email@example.com|
|Version:||trunk||Keywords:||ossman_tester, relnotes, samuel_tester|
We call netstat once for each session, even though the information is relevant for all sessions. IOW we could optimise things to one invocation per XMLRPC call.
After the work done on bug 5489 all netstat calls are run within the same timeout. We have decided that we will take a look at this as a continuation of that bug.
For each session to verify, we first makes a call to ps for validating that the process is an tl-session / xinit process. Then we make a call to netstat to verify if a client is connected to the session. We could do the checks in one call for each subprocess (ps, netstat). First a call to ps wtih a set of sessions to verify, failed sessions are added to result and removed from the set of sessions. Next pass being running netstat with the remaining sessions in the the set, adding them to result as succesfully verified sessions.
Another approach for optimizing the ps part is to read the link /proc/%d/exe to verify which binary is running as pid %d.
Looks good. Verified using ptrace that only a single invocation of 'ss' is made, no matter the number of sessions. It also updates the connection status properly.
The use of any() is limited to Python 2.5 and newer. Tracebacks on RHEL5 (with Python 2.4): > 2017-02-06 11:20:48 ERROR vsmagent: Unhandled exception with async process: exceptions.NameError global name 'any' is not defined Traceback (most recent call last): > File "/opt/thinlinc/modules/thinlinc/vsm/async.py", line 103, in i1Iii > obj . handle_read_event ( fd ) > File "/opt/thinlinc/modules/thinlinc/vsm/extproc.py", line 311, in handle_read_event > self . handle_close ( fd ) > File "/opt/thinlinc/modules/thinlinc/vsm/extproc.py", line 284, in handle_close > self . handle_exit ( ) > File "/opt/thinlinc/modules/thinlinc/vsm/extproc.py", line 457, in handle_exit > self . callback ( O00oO000O0O , self . stdout_data , self . stderr_data ) > File "/opt/thinlinc/modules/thinlinc/vsm/handler_verifysessions.py", line 28, in <lambda> > o00 = lambda Oo0oO0ooo , o0oOoO00o , i1 : self . _parse_ss_output ( Oo0oO0ooo , o0oOoO00o , i1 , IiiIII111iI ) > File "/opt/thinlinc/modules/thinlinc/vsm/handler_verifysessions.py", line 125, in _parse_ss_output > self . _process_sessions ( sessions , Ii1IOo0o0 ) > File "/opt/thinlinc/modules/thinlinc/vsm/handler_verifysessions.py", line 88, in _process_sessions > if any ( [ oo . endswith ( ":%d" % IiiiI1II1I1 ) for oo in listeners ] ) : > NameError: global name 'any' is not defined
Works now. Retested on RHEL 5 and RHEL 6.
Upgrading from 4.7.0 on RHEL 5 loses sessions, and it seems to be caused by the code changed in this bug: > 2017-05-18 10:49:15 WARNING vsmagent.session: Broken session for user cendio, tl-session pid 4869 is not tl-session > [root@dhcp-253-137 tl-4.8.0-server]# ps fax| grep 4869 > 6401 pts/1 S+ 0:00 | \_ grep 4869 > 4869 ? S 0:00 tl-session: cendio
It seems like older kernels behave slightly different for deleted files. This is the contents of the 'exe' symlink: /opt/thinlinc/libexec/tl-session\x00591d636f (deleted) The (illegal) null in there causes most tools to not display the trailing stuff. Unfortunately strings aren't null terminated in Python so it gets confused by this.
Tested on RHEL 5 and reconnect works well both for upgraded and new session.
Tested on Ubuntu 16.04 and reconnect works well both for upgraded and new session.