[0.10.0] panic in model.Points.scanTo

[dvusboy]

I'm using telegraf feeding into influxdb. In order to compute the % CPU usage of a container as a time series, it is necessary to compute the change in measurements like docker_cpu.total_usage. Since expressions of aggregates and the function difference() aren't supported yet, I thought of using a continuous query to precompute these deltas. Basically, within a GROUP BY time interval, compute LAST(total_usage) - FIRST(total_usage) (or MAX(total_usage) - MIN(total_usage), they should be the same).

What I'm running into is a panic and influxdb crashes.

• Is my query messed up? Or,
• Is this a bug in influxdb?

I've included the backtrace and the query in this gist.

[jwilder]

@e-dard 789aff9 might have introduced this. Any ideas?

[e-dard]

@jwilder fixed this. Though it's slightly perplexing because scanTo only gets called for various delimiter values, e.g., = or , AFAIK. In the case that the stop value is also the first character of buf, the parser should error out before a call to scanTo. I will revisit in the future.

[dvusboy]

Alas, I don't have the full log leading up to the crash. All I have is this snippet just before the backtrace in the gist:

2016-02-12T21:26:55.676264820Z [meta] 2016/02/12 21:26:55 127.0.0.1 - - [12/Feb/2016:21:26:55 +0000] GET /lease?name=continuous_querier&nodeid=0 HTTP/1.1 200 106 - Go-http-client/1.1 573d5b0f-d1cf-11e5-803c-000000000000 1.781658ms
2016-02-12T21:26:55.873619472Z [wal] 2016/02/12 21:26:55 write to index of partition 1 took 245.31988ms
2016-02-12T21:26:56.679888889Z [meta] 2016/02/12 21:26:56 127.0.0.1 - - [12/Feb/2016:21:26:56 +0000] GET /lease?name=continuous_querier&nodeid=0 HTTP/1.1 200 106 - Go-http-client/1.1 57d68199-d1cf-11e5-803d-000000000000 1.7249ms
2016-02-12T21:26:57.683457109Z [meta] 2016/02/12 21:26:57 127.0.0.1 - - [12/Feb/2016:21:26:57 +0000] GET /lease?name=continuous_querier&nodeid=0 HTTP/1.1 200 106 - Go-http-client/1.1 586fa3ce-d1cf-11e5-803e-000000000000 1.745397ms
2016-02-12T21:26:58.687429516Z [meta] 2016/02/12 21:26:58 127.0.0.1 - - [12/Feb/2016:21:26:58 +0000] GET /lease?name=continuous_querier&nodeid=0 HTTP/1.1 200 106 - Go-http-client/1.1 5908c589-d1cf-11e5-803f-000000000000 2.017598ms
2016-02-12T21:26:59.691900968Z [meta] 2016/02/12 21:26:59 127.0.0.1 - - [12/Feb/2016:21:26:59 +0000] GET /lease?name=continuous_querier&nodeid=0 HTTP/1.1 200 106 - Go-http-client/1.1 59a218c8-d1cf-11e5-8040-000000000000 1.740024ms
2016-02-12T21:27:00.297889443Z [http] 2016/02/12 21:27:00 192.168.22.99 - - [12/Feb/2016:21:27:00 +0000] POST /write?consistency=&db=telegraf&precision=s&rp= HTTP/1.1 204 0 - InfluxDBClient 59f0bd44-d1cf-11e5-8041-000000000000 92.326954ms
2016-02-12T21:27:00.695455927Z [meta] 2016/02/12 21:27:00 127.0.0.1 - - [12/Feb/2016:21:27:00 +0000] GET /lease?name=continuous_querier&nodeid=0 HTTP/1.1 200 106 - Go-http-client/1.1 5a3b3b50-d1cf-11e5-8042-000000000000 1.736889ms

I hope this helps.

[jonseymour]

@e-dard just in case it is useful to you, I have a query that caused this in my database.

select count(value)/60 into "default"."count.foobar" from "foobar.state" where time > now() - 24h group by time(5m), foo, bar

The series "foobar.state" has to exist and have data in the range specified by the time filter - it won't crash otherwise.