By the time the aggregator.run() was called about 600ms already passed since setting now which was skewing up the aggregation intervals and skipping metrics.
Looping the metrics back through the same channel could result in a
deadlock, by using a new channel and locking the processor we can ensure
that all stages can make continual progress.
doing this unblocks incoming metrics while waiting for a flush to take
place.
we have to create a semaphore so that we can
'skip' flushes that try to run while a flush is already running.
closes#2262
in this commit:
- centralize logging output handler.
- set global Info/Debug/Error log levels based on config file or flags.
- remove per-plugin debug arg handling.
- add a I!, D!, or E! to every log message.
- add configuration option to specify where to send logs.
closes#1786
This is for better thread-safety when running with multiple outputs,
which can cause very odd panics at very high loads
primarily this is to address #1432closes#1432
Adding precision rounding to the accumulator. This means that now every
input metric will get rounded at collection, rather than at write (and
only for the influxdb output).
This feature is disabled for service inputs, because service inputs
should be in control of their own timestamps & precisions.
use a common function between collection_jitter and flush_jitter. which
creates the same behavior between the two options.
going forward, both jitters will be random sleeps that get re-evaluated
at runtime for every interval (previously only collection_jitter did
this)
also fixes behavior so that both jitters will exit in the event of a
process exit.
closes#1296
Also changing the net_response and http_response plugins to only accept
duration strings for their timeout parameters. This is a breaking config
file change.
closes#1214
Changing the internal behavior around running plugins. Each plugin
will now have it's own goroutine with it's own ticker. This means that a
hung plugin will not block any other plugins. When a plugin is hung, we
will log an error message every interval, letting users know which
plugin is hung.
Currently the input interface does not have any methods for killing a
running Gather call, so there is nothing we can do but log an "ERROR"
and move on. This will give some visibility into the plugin that is
acting up.
closes#1230fixes#479
Steven Soroka
dbutler-starry
Daniel Nelson
Greg
Frank Riley
Daniel Nelson
John Hu
Adrián López
Piotr Popieluch
Maximilien Richer
Cameron Sparr
Cosmo Petrich
Patrick Hemmer