Updated to 0.8.8b - suddenly get SNMP timeouts?

[Howie]

The log entries I posted above were from an snmp ver 2c device.

So we're agreed that version makes no difference

[mozolins]

I'm glad we agree, what do we do about it?

[gandalf]

I'm glad we agree, what do we do about it?

Perform a diff of the source files and find out, what has been changed. Perhaps additionally use the Changelog as a hint.
Or scan the SVN updates between those versions.
Only I do not find time right now to do this on my own.

Yes, a bit of C knownledge will help ...
R.

[mozolins]

thanks for your support.

I'm somewhat busy myself, finding time to find the misplaced ; wasn't in this weeks plan.

[mozolins]

This might be part of the problem. this is a diff from 08.7 just before the tag for 0.8.8. It seems a number of header files are no longer included in the program.
spine/branches/0.8.7/snmp.c revision 6053, Thu Jul 15 03:23:08 2010 UTC spine/branches/0.8.8/snmp.c revision 7398, Sun Aug 4 20:32:56 2013 UTC

Code: Select all

	Line 34 	Line 34 
34	 #include "common.h"	 #include "common.h"
35	 #include "spine.h"	 #include "spine.h"
36	 	 
 #ifdef USE_NET_SNMP	 
  #undef PACKAGE_NAME	 
  #undef PACKAGE_VERSION	 
  #undef PACKAGE_BUGREPORT	 
  #undef PACKAGE_STRING	 
  #undef PACKAGE_TARNAME	 
  #include <net-snmp/net-snmp-config.h>	 
  #include <net-snmp/utilities.h>	 
  #include <net-snmp/net-snmp-includes.h>	 
  #include <net-snmp/config_api.h>	 
  #include <net-snmp/mib_api.h>	 
 #else	 
  #include <ucd-snmp/ucd-snmp-config.h>	 
  #include <ucd-snmp/ucd-snmp-includes.h>	 
  #include <ucd-snmp/system.h>	 
  #include <mib.h>	 
 #endif	 
 	 
37	 /* resolve problems in debian */	 /* resolve problems in debian */
38	 #ifndef NETSNMP_DS_LIB_DONT_PERSIST_STATE	 #ifndef NETSNMP_DS_LIB_DONT_PERSIST_STATE
39	  #define NETSNMP_DS_LIB_DONT_PERSIST_STATE 32	  #define NETSNMP_DS_LIB_DONT_PERSIST_STATE 32

[Howie]

At the time, I just stuck with my old version of spine, but I'd like to get up to date - was this resolved in 0.8.8c? Nothing in the changelog mentions timeouts...

[Howie]

Hmm, I still see similar issues in Spine 0.8.8c

Code: Select all

# /usr/local/spine/bin/spine -C /usr/local/spine/etc/spine.conf -R -H 141

[snip]
04/15/2015 10:27:39 AM - SPINE: Poller[0] Host[141] TH[1] DS[3991] WARNING: SNMP timeout detected [4500 ms], ignoring host '10.0.0.21'
04/15/2015 10:27:39 AM - SPINE: Poller[0] Host[141] TH[1] DS[3991] WARNING: SNMP timeout detected [4500 ms], ignoring host '10.0.0.21'
[snip]

(and still, the entire run of about 50 DS takes a fraction of a second, not 4.5)

[Howie]

So it seems that the error message is not very accurate. *ANY* issue with setting up the SNMP session is reported as a timeout.

Code: Select all

 /* catch snmp initialization issues */
                                if (!host->snmp_session) {
                                        host->ignore_host = TRUE;
                                        break;
                                }

[.....]

 if (host->ignore_host) {
            SPINE_LOG(("Host[%i] TH[%i] DS[%i] WARNING: SNMP timeout detected [%i ms], ignoring host '%s'", host_id, host_thread, poller_items[snmp_oids[j].array_position].local_data_id, host->snmp_timeout, host->hostname));

[Howie]

So after building a debug version of net-snmp and a test version of Spine, I got to this trace message just before things start to fail:

sess_async_send: length of packet (1763) exceeds session maximum (1500)

I have changed the maximum OIDs per request to 10, and now I'm getting data! Nothing to do with timeouts, as expected.