[Net-snmp-patches] [ net-snmp-Patches-919122 ] test should allow time for programs to settle

SourceForge Headquarters 225 Broadway Suite 1600 San Diego, CA 92101 +1 (858) 454-5900

Patches item #919122, was opened at 2004-03-18 17:33
Message generated for change (Comment added) made by slif
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=312694&aid=919122&group_id=12694

Category: None
Group: None
Status: Open
Resolution: None
Priority: 5
Submitted By: Michael J. Slifcak (slif)
Assigned to: Nobody/Anonymous (nobody)
Summary: test should allow time for programs to settle

Initial Comment:
The agent and the trap receiver must perform extensive
initialization before they are ready to perform their
intended function.  

[The next statement is true for net-snmp-5.x through
version 5.1.1]
When the agent is configured as an agentX subagent,
it will transact 20 packets (10 send, 10 receive)
before it will print the version string.

Why then, should the test framework expect to not wait
SNMP_SLEEP seconds before looking for that version string ?

Apply this patch to wait after starting the agent or
the trap receiver, no matter how either is started.

----------------------------------------------------------------------

>Comment By: Michael J. Slifcak (slif)
Date: 2004-03-29 13:43

Message:
Logged In: YES 
user_id=88697

patch-test-settle-6 
The delay after start is replaced by a more generous repeat
cycle
in the WAITFOR function.  Here the time to wait for expected
result
is expanded to five times the SNMP_SLEEP setting.

For the default of SNMP_SLEEP=1, this creates a five second
window
during which the agent (or trap receiver) can respond.  Fast
responders
are rewarded by exiting the loop early.  Slow responders are
given
more time to produce the expected result.

Given this approach, there is no longer any need to DELAY after
starting or stopping a program.  Since the NOSLEEP methods are 
found only in eval_tools.sh, they have been replaced with
STARTPROG or STOPPROG, as appropriate.

The NOSLEEP methods are removed by this patch, as they are
not referenced elsewhere.

----------------------------------------------------------------------

Comment By: Michael J. Slifcak (slif)
Date: 2004-03-29 12:52

Message:
Logged In: YES 
user_id=88697

Previous comments from cfmartin do not apply here.

patch-test-settle-4 is not platform specific
It will repeat the search for WAITFOR''s input string
for up to 5 seconds.   It works without explicitly testing
for output file existence.  It does not force a delay
for platforms that have agents which respond qucikly.

Since WAITFOR is called after starting and stopping programs,
the repeating for 5 seconds feature might yield acceptable
results
on AIX.   This theorem is not verified.

----------------------------------------------------------------------

Comment By: ProudlyBrewStarbucks (cfmartin)
Date: 2004-03-25 14:19

Message:
Logged In: YES 
user_id=1005866

I see now that the patch was for the RUNTESTS script, which 
runs successfully without the patch.

----------------------------------------------------------------------

Comment By: ProudlyBrewStarbucks (cfmartin)
Date: 2004-03-25 13:31

Message:
Logged In: YES 
user_id=1005866

I just opened 922835 "snmptrapd cannot be started with
init".  I was wondering if this is related to the problem
here that snmptrapd does not have enough time to initialize.  

The following is an example of what gets written to
/var/adm/messages when init tries to start snmptrapd.  This
problem is also in 5.1.1.  Port 1622 is definitely not in
use before attempting to start snmptrapd.

I have not applied a patch before.  Is there some doc on how
to  go about it?   Thanks.

Mar 24 17:22:17 dsm-nt-clnt2 syslog[17078]: [ID 702911
daemon.error] pdu failed to be created
Mar 24 17:22:17 dsm-nt-clnt2 last message repeated 654 times
Mar 24 17:22:18 dsm-nt-clnt2 syslog[17080]: [ID 702911
daemon.error] couldn't open 1622 -- errno 125 ("Address
already in use")
Mar 24 17:22:18 dsm-nt-clnt2 syslog[17078]: [ID 702911
daemon.error] pdu failed to be created
Mar 24 17:22:18 dsm-nt-clnt2 last message repeated 540 times
Mar 24 17:22:19 dsm-nt-clnt2 syslog[17082]: [ID 702911
daemon.error] couldn't open 1622 -- errno 125 ("Address
already in use")
Mar 24 17:22:19 dsm-nt-clnt2 syslog[17078]: [ID 702911
daemon.error] pdu failed to be created
Mar 24 17:22:19 dsm-nt-clnt2 last message repeated 528 times
Mar 24 17:22:20 dsm-nt-clnt2 syslog[17084]: [ID 702911
daemon.error] couldn't open 1622 -- errno 125 ("Address
already in use")
Mar 24 17:22:20 dsm-nt-clnt2 syslog[17078]: [ID 702911
daemon.error] pdu failed to be created
Mar 24 17:22:20 dsm-nt-clnt2 last message repeated 389 times
Mar 24 17:22:20 dsm-nt-clnt2 syslog[17086]: [ID 702911
daemon.error] couldn't open 1622 -- errno 125 ("Address
already in use")
Mar 24 17:22:20 dsm-nt-clnt2 syslog[17078]: [ID 702911
daemon.error] pdu failed to be created
Mar 24 17:22:21 dsm-nt-clnt2 last message repeated 638 times
Mar 24 17:22:21 dsm-nt-clnt2 syslog[17088]: [ID 702911
daemon.error] couldn't open 1622 -- errno 125 ("Address
already in use")
Mar 24 17:22:22 dsm-nt-clnt2 syslog[17078]: [ID 702911
daemon.error] pdu failed to be created
Mar 24 17:22:23 dsm-nt-clnt2 last message repeated 1242 times
Mar 24 17:22:23 dsm-nt-clnt2 syslog[17090]: [ID 702911
daemon.error] couldn't open 1622 -- errno 125 ("Address
already in use")
Mar 24 17:22:23 dsm-nt-clnt2 syslog[17078]: [ID 702911
daemon.error] pdu failed to be created
Mar 24 17:22:24 dsm-nt-clnt2 last message repeated 310 times
Mar 24 17:22:24 dsm-nt-clnt2 syslog[17092]: [ID 702911
daemon.error] couldn't open 1622 -- errno 125 ("Address
already in use")
Mar 24 17:22:24 dsm-nt-clnt2 syslog[17078]: [ID 702911
daemon.error] pdu failed to be created
Mar 24 17:22:25 dsm-nt-clnt2 last message repeated 419 times
Mar 24 17:22:25 dsm-nt-clnt2 syslog[17094]: [ID 702911
daemon.error] couldn't open 1622 -- errno 125 ("Address
already in use")
Mar 24 17:22:25 dsm-nt-clnt2 syslog[17078]: [ID 702911
daemon.error] pdu failed to be created
Mar 24 17:22:26 dsm-nt-clnt2 last message repeated 360 times
Mar 24 17:22:26 dsm-nt-clnt2 syslog[17096]: [ID 702911
daemon.error] couldn't open 1622 -- errno 125 ("Address
already in use")
Mar 24 17:22:26 dsm-nt-clnt2 syslog[17078]: [ID 702911
daemon.error] pdu failed to be created
Mar 24 17:22:27 dsm-nt-clnt2 last message repeated 556 times
Mar 24 17:22:27 dsm-nt-clnt2 syslog[17098]: [ID 702911
daemon.error] couldn't open 1622 -- errno 125 ("Address
already in use")
Mar 24 17:22:27 dsm-nt-clnt2 syslog[17078]: [ID 702911
daemon.error] pdu failed to be created

----------------------------------------------------------------------

Comment By: Michael J. Slifcak (slif)
Date: 2004-03-20 13:34

Message:
Logged In: YES 
user_id=88697

I understand the logic of not sleeping SLEEP seconds for
most platforms.
That works reliably only when the agent or trap receiver are
started so that
they exit after having forked and begun initializing
themselves as
UNIX daemons, before the shell comes through the WAITFOR logic
to grep messages in the logfile.

On mingw, or with MSVS builds, there is no fork. The agent
and trap receiver
are started from background (&).  But I believe that adding
a platform
specific sleep would not offer robust execution everywhere,
for three reasons:

1. If the logfile did not exist, the grep would fail.
2. If the logfile did not already have the string, the grep
would fail.

3. If the agent started as an AgentX subagent, it passes 10
packets
to and 10 packets from the master agent before it writes the
version string.   The grep may fail here as well.

I'll concede that the first patch did not perform a globally
useful
function.

If WAITFORNOSLEEP were changed to always wait for 2 seconds,
that would force the shell to yield, and may give enough
time for
the agent to complete its initializations and print its
version strings.

The second patch does this.

----------------------------------------------------------------------

Comment By: Wes Hardaker (hardaker)
Date: 2004-03-19 18:52

Message:
Logged In: YES 
user_id=76242

not applying for 5.1.1.  The WAITFOR line should take care of a 
need ffor sleep.  the whole point is that it will wait till the version 
string line shows up, which is a better way for determining if the 
agent is truely up than a hard sleep.  thoughts? 

----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=312694&aid=919122&group_id=12694