influxdb: Concurrent read- and write-request leads to error "query interrupted" on 1.4

System info: InfluxDB 1.4.2 and 1.4.1 on debian linux

Steps to reproduce:

Run concurrently many query- and write-requests againts the http endpoint. I was able to reproduce the problem with 3 parallel treads doing SELECT-Queries and 2 parallel threads doing writes against the same set of series.

Unfortunatly I can’t easily provide a shell script to reproduce, because I’m running a windows .net-application as http client. But I will provide a wireshark file with the observed behaviour.

Expected behavior:

All queries should get an proper result. If server-error occur they should be logged in the journal and counted in the _internal,http measurement.

Actual behavior:

Some of the queries get an HTTP-200 result but in the response there is the following json string: {"results":[{"statement_id":0, "error":"query interrupted"}]}

See the following packet-numbers in the attached wirehark-file:

• 204
• 430
• 1797

Additional info:

If I run the same (or even a much higher) load against a 1.3.7 influxdb on the same machine, all queries return the expected result. I also coundn’t see any impact on changing the max-concurrent-queries parameter in the coordinator section of the influxdb.config. During the time I recorded the wireshark-file this parameter was set to 0.

You should easily be able to reproduce this creating similar load as you can see in the wireshark-file. If you can’t, let me know and I’ll provide additional information for you. Influxdb-quer_interrupted.pcapng.gz