fluent-plugin-numeric-counter

Component

NumericCounterOutput

Fluentd plugin to count messages, matches for numeric range patterns, and emits its result (like fluent-plugin-datacounter).

Counts per min/hour/day
Counts per second (average every min/hour/day)
Percentage of each numeric pattern in total counts of messages

NumericCounterOutput emits messages contains results data, so you can output these message (with 'numcount' tag by default) to any outputs you want.

output ex1 (aggregates all inputs): {"pattern1_count":20, "pattern1_rate":0.333, "pattern1_percentage":25.0, "pattern2_count":40, "pattern2_rate":0.666, "pattern2_percentage":50.0, "unmatched_count":20, "unmatched_rate":0.333, "unmatched_percentage":25.0}
output ex2 (aggregates per tag): {"test_pattern1_count":10, "test_pattern1_rate":0.333, "test_pattern1_percentage":25.0, "test_pattern2_count":40, "test_pattern2_rate":0.666, "test_pattern2_percentage":50.0, "test_unmatched_count":20, "test_unmatched_rate":0.333, "test_unmatched_percentage":25.0}

'input_tag_remove_prefix' option available if you want to remove tag prefix from output field names.

If you want to omit 'unmatched' messages from percentage counting, specify 'outcast_unmatched yes'.

Configuration

NumericCounterOutput

Count messages that have attribute 'duration'(response time by microseconds), by several numeric ranges, per minutes.

<match accesslog.**>
  type numeric_counter
  unit minute           # or 'count_interval 60s' or '45s', '3m' ... as you want
  aggregate all         # or 'tag'
  count_key duration

  # patternX: X(1-20)
  # patternX NAME LOW HIGH  #=> patternX matches N like LOW <= N < HIGH

  pattern1 HIGHSPEED            0 10000 # under 10ms
  pattern2 SEMIHIGHSPEED   10000 100000 # under 100ms
  pattern3 NORMAL        100000 1000000 # under 1s
  pattern4 STUPID       1000000 10000000 # under 10s!

  # patternZ (Z is last number of specified patterns)
  # patternZ NAME LOW  #=> patternZ matches N like LOW <= N (upper threshold is unlimited)
  patternZ MUSTDIE     10000000  # over 10s!
</match>

Size specifier (like 10k, 5M, 103g) available as 1024**1, 1024**2, 1024**3 ...., for example, for bytes of access log.

<match accesslog.**>
  type numeric_counter
  unit hour
  aggregate tag
  count_key bytes

  pattern1 SMALL    0 1k
  pattern2 MIDDLE  1k 1m
  pattern3 LARGE   1m 10m
  pattern4 HUGE   10m 1g
  pattern5 XXXX    1g
</match>

You can try to use negative numbers, and floating point numbers.... (not tested enough).

With 'output_per_tag' option and 'tag_prefix', we get one result message for one tag:

<match accesslog.{foo,bar}>
  type numeric_counter
  unit hour
  aggregate tag
  count_key bytes
  output_per_tag yes
  tag_prefix byteslog
  input_tag_remove_prefix accesslog

  pattern1 SMALL    0 1k
  pattern2 MIDDLE  1k 1m
  pattern3 LARGE   1m 10m
  pattern4 HUGE   10m 1g
  pattern5 XXXX    1g
</match>
# => tag: 'byteslog.foo' and 'byteslog.bar'
#    message: {'SMALL_count' => 100, ... }

And you can get tested messages count with 'output_messages' option:

<match accesslog.{foo,bar}>
  type numeric_counter
  unit hour
  aggregate tag
  count_key bytes
  input_tag_remove_prefix accesslog
  output_messages yes

  pattern1 SMALL    0 1k
  pattern2 LARGE   1k
</match>
# => tag: 'numcount'
#    message: {'foo_messages' => xxx, 'bar_messages' => yyy, 'foo_SMALL_count' => 100, ... }

<match accesslog.{foo,bar}>
  type numeric_counter
  unit hour
  aggregate tag
  count_key bytes
  output_per_tag yes
  tag_prefix num
  input_tag_remove_prefix accesslog
  output_messages yes

  pattern1 SMALL    0 1k
  pattern2 LARGE   1k
</match>
# => tag: 'num.foo' or 'num.bar'
#    message: {'messages' => xxx, 'SMALL_count' => 100, ... }

Parameters

count_key (required)

The key to count in the event record.
tag

The output tag. Default is numcount.
tag_prefix

The prefix string which will be added to the input tag. output_per_tag yes must be specified together.
input_tag_remove_prefix

The prefix string which will be removed from the input tag.
count_interval

The interval time to count in seconds. Default is 60.
unit

The interval time to monitor specified an unit (either of minute, hour, or day). Use either of count_interval or unit.
aggregate

Calculate in each input tag separetely, or all records in a mass. Default is tag.
ouput_per_tag

Emit for each input tag. tag_prefix must be specified together. Default is no.
outcast_unmatched

Specify yes if you do not want to include 'unmatched' counts into percentage. Default is no.
output_messages

Specify yes if you want to get tested messages. Default is no.
store_file

Store internal data into a file of the given path on shutdown, and load on starting.

TODO

more tests
more documents

Copyright

License
- Apache License, Version 2.0