Telegraf inputs.http_listener ignores timestamp precision

[Menthalion]

Bug report

When Telegraf with http_listener is used as a proxy for InfluxDB, the http request parameter precision=<unit> is ignored, and times are passed through as nanoseconds even if they aren't.

Relevant telegraf.conf:

Default inputs.http_listener and either http or udp outputs.influxdb

System info:

Telegraf 1.1.2, InfluxDB 1.2 on Ubuntu 16.04

Steps to reproduce:

1. Set up a Telegraf with inputs.http_listener that forwards it's data to Influxdb
2. Send a http request to Telegraf with precision=s and timestamps in seconds/epoch
3. Query InfluxDb to see the timestamp

Expected behavior:

The timestamp in influxdb is registered in seconds / epoch, and corresponds with the time of the measuremt

Actual behavior:

The timestamp is registered in nanoseconds / epoch, and correspond with a time in 1970

Proposal:

Either have Telegraf cast the timestamp from the input precision to nanoseconds, or pass through both timestamp and precision to the outputs.

Use case:

Most metrics systems supporting InfluxDB only support HTTP requests, but a high load of HTTP requests cause performance problems by each causing a WAL sync.

So that's why we need either

• Telegraf as dependable HTTP->Telegraf->UDP->InfluxDB proxy, or
• InfluxDB needs to be adjusted so HTTP requests will be buffered for an amount points before a write just like the UDP / TCP / Graphite inputs do. This would mean the status returned on such a http request should be a 202 (Accepted). An additional parameter for guaranteed write could also be created.

[sparrc]

telegraf is not meant to be a proxy for InfluxDB, it accepts line-protocol which is defined as having nanosecond timestamps.