[SmartOS] Broken collectd-network in 2016Q4

[hadret]

Hi,

recently I noticed that nodes I migrated to base-64-lts 16.4.1 are no longer sending metrics to my central graphite server. Turned out that collectd daemon was failing to start due to fatal error. Here's the message:

[ Apr 13 11:03:47 Enabled. ]
[ Apr 13 11:03:47 Executing start method ("/opt/local/sbin/collectd"). ]
configfile: stat (/opt/local/etc/collectd.d/*.conf) failed: No such file or directory
ld.so.1: collectd: fatal: relocation error: file /opt/local/lib/collectd/network.so: symbol gcry_control: referenced symbol not found
[ Apr 13 11:03:47 Method "start" failed due to signal KILL. ]

image:base-64-lts 16.4.1
collectd:5.6.2
platform:joyent_20161222T003450Z

Regards,
Filip

[mamash]

Hm, it's strange. I see the network.so lib doesn't link libgcrypt like it should (I think), but I can't bring collectd crashing on the missing symbol, it just keeps running. I'll continue investigating.

[hadret]

Ah, probably I should have include also the configuration I'm using! Here's the relevant part:

LoadPlugin network
<Plugin network>
        <Server "$REMOTE_SERVER" "$REMOTE_PORT">
                SecurityLevel Sign
                Username "$USER"
                Password "$PASSWORD"
        </Server>
        MaxPacketSize 1500
</Plugin>

If link against libgcrypt should be there and it is part responsible for encryption in network plugin, then this might be the very reason it fails.

[mamash]

Ah OK, confirmed.

[mamash]

I fixed this in pkgsrc trunk, will backport to 2016Q4.

[hadret]

Great! Many thanks!

[mamash]

Should be fixed in 2016Q4, I backported the update to 5.7.1 too.