[1.6] "no fields" error when doing post to influxdb

[albundy83]

Bug report

After upgrading to 1.6.0 release, cassandra metrics are not sent anymore.
I have the following error messages:

2018-04-18T14:14:10Z E! [outputs.influxdb]: when writing to [http:/my_influxdb:8086]: Post http://my_influxdb:8086/write?consistency=any&db=telegraf: no fields

Relevant telegraf.conf:

We use jolokia and cassandra to download JMX informations as written in:

https://github.com/influxdata/telegraf/tree/master/plugins/inputs/cassandra

The configuration file is quite simple, the part that is changed is on cassandra inputs:
(it's just an extract)

[[inputs.cassandra]]
  context = "/jolokia/read"
  servers = ["localhost:8778"]
  metrics = [
    "/java.lang:type=Memory/HeapMemoryUsage/used",
    "/java.lang:type=Memory/NonHeapMemoryUsage/used",
    "/java.lang:type=GarbageCollector,name=ConcurrentMarkSweep/CollectionTime",
    "/java.lang:type=GarbageCollector,name=ConcurrentMarkSweep/CollectionCount",
    "/java.lang:type=GarbageCollector,name=ParNew/CollectionTime",
    "/java.lang:type=GarbageCollector,name=ParNew/CollectionCount",
    "/org.apache.cassandra.metrics:type=Cache,scope=CounterCache,name=Capacity",
    "/org.apache.cassandra.metrics:type=Cache,scope=CounterCache,name=Entries",
    "/org.apache.cassandra.metrics:type=Cache,scope=CounterCache,name=HitRate",
    "/org.apache.cassandra.metrics:type=Cache,scope=CounterCache,name=Hits",
    "/org.apache.cassandra.metrics:type=Cache,scope=CounterCache,name=Requests",
    "/org.apache.cassandra.metrics:type=Cache,scope=CounterCache,name=Size",
    "/org.apache.cassandra.metrics:type=Cache,scope=KeyCache,name=Capacity",
    "/org.apache.cassandra.metrics:type=Cache,scope=KeyCache,name=Entries",
    "/org.apache.cassandra.metrics:type=Cache,scope=KeyCache,name=HitRate"
 ]

System info:

Telegraf 1.6.0
Tested on 2 Ubuntu 14.04 with the following kernels:
4.4.0-112-generic
3.13.0-137-generic

Steps to reproduce:

Upgrade to telegraf 1.6.0
Check that you don't have any cassandra metrics
See the error messages:

2018-04-18T14:14:10Z E! [outputs.influxdb]: when writing to [http:/my_influxdb:8086]: Post http://my_influxdb:8086/write?consistency=any&db=telegraf: no fields

Rollback to telegraf 1.5.3
Metrics are back

In debug mode, I don't have more details.

[albundy83]

On influxdb side, I have the following errors:

Apr 18 16:56:33 my_influxdb influxd[7503]: [httpd] 10.234.151.183 - telegraf [18/Apr/2018:16:56:33 +0200] "POST /write?consistency=any&db=telegraf HTTP/1.1" 400 27 "-" "telegraf" af58b217-4318-11e8-a196-000000000000 9645
Apr 18 16:56:36 my_influxdb influxd[7503]: [httpd] 10.234.151.183 - telegraf [18/Apr/2018:16:56:36 +0200] "POST /write?consistency=any&db=telegraf HTTP/1.1" 400 27 "-" "telegraf" b1461478-4318-11e8-a1af-000000000000 585

[danielnelson]

Thank you for the report, could you run telegraf --input-filter cassandra --test in 1.5.3 and add the output?

[albundy83]

Here the output with telegraf 1.5.3:
https://gist.github.com/albundy83/b0fb9484ba666f68112013811cdc0a21

[danielnelson]

Lots of "interesting" metrics in there. I see a lot of map conversions that I know are going to be removed in 1.6, since I'm assuming you are not interested in storing measurements with fields like this:

Hits_RateUnit="map[declaringClass:map[interfaces:[] name:java.util.concurrent.TimeUnit]]"

I'm still having a hard time tracking down the error sending to InfluxDB though, could you also run that same command with 1.6.0? Also, you mentioned that the cassandra config was an extract, do you have more metrics specified and if so can you add the full config for the cassandra plugin?

[albundy83]

Here the same output with telegraf 1.6.0:
https://gist.github.com/albundy83/a09c99780c7fb514d866f565643211c3

And here the config we use with cassandra:
https://gist.github.com/albundy83/f3d86f130d9aa38f92c2c78c2c557502

[albundy83]

And yes, the measurements with java classes etc ... are gone in 1.6, here a comparison:
in 1.5.3

cassandraCache,scope=CounterCache,mname=Hits,cassandra_host=localhost,rack=rack1,env=prod,anneau=cluster01,host=my_hostname.com,dc=dc1 Hits_Count=0,Hits_FifteenMinuteRate=0,Hits_FiveMinuteRate=0,Hits_MeanRate=0,Hits_RateUnit="map[declaringClass:map[interfaces:[] name:java.util.concurrent.TimeUnit]]",Hits_OneMinuteRate=0,Hits_EventType="hits" 1524120274000000000

in 1.6.0

cassandraCache,anneau=cluster01,cassandra_host=localhost,dc=dc1,env=prod,host=my_hostname.com,mname=Hits,rack=rack1,scope=CounterCache Hits_FifteenMinuteRate=0,Hits_FiveMinuteRate=0,Hits_MeanRate=0,Hits_OneMinuteRate=0,Hits_EventType="hits",Hits_Count=0 152412190800000000

[danielnelson]

Thank you, with the files you provided I was able to reproduce the bug, so it shouldn't be a problem to put together a fix.

Just to clarify, are you okay if these fields with java classes are gone? I believe their inclusion was a bug.

[albundy83]

Yes sure, we can't do nothing with it :-)

[albundy83]

Can you give me some information about the issue if you have time please ?

[danielnelson]

The error is caused by the metric having no fields, which is caused by the removal of these junk fields. When we serialize it to line protocol we return this error, and we are failing the entire batch because of it. What complicates it in this case is that we are serializing while sending the HTTP request to InfluxDB, instead of doing it all in memory before making the request. This causes the error to look as if it may be a HTTP issue, when actually it is the serializer.