New metric system (Prometheus data format)

[MrAnno]

This PR adds a new metric naming scheme (labeling capability) to syslog-ng and a querying tool to produce Prometheus-like metric exports.

$ sbin/syslog-ng-ctl stats prometheus

syslogng_events_allocated_bytes 18446744073709550560
syslogng_filtered_events_total{id="#anon-filter0",result="matched"} 0
syslogng_filtered_events_total{id="#anon-filter0",result="not_matched"} 2
syslogng_filtered_events_total{id="ff",result="matched"} 0
syslogng_filtered_events_total{id="ff",result="not_matched"} 2
syslogng_input_events_total{id="#anon-source0#0",driver_instance="-",result="processed"} 0
syslogng_input_events_total{id="s_network#1",result="processed"} 2
syslogng_internal_source{result="dropped"} 0
syslogng_internal_source{result="queued"} 0
syslogng_output_events_total{id="d_dest#0",driver_instance="tcp,127.0.0.1:5555",result="delivered"} 0
syslogng_output_events_total{id="d_dest#0",driver_instance="tcp,127.0.0.1:5555",result="dropped"} 0
syslogng_output_events_total{id="d_dest#0",driver_instance="tcp,127.0.0.1:5555",result="queued"} 0
syslogng_output_events_total{id="d_dest#1",driver_instance="http,https://localhost:8080",result="delivered"} 0
syslogng_output_events_total{id="d_dest#1",driver_instance="http,https://localhost:8080",result="dropped"} 0
syslogng_output_events_total{id="d_dest#1",driver_instance="http,https://localhost:8080",result="queued"} 0
syslogng_parsed_events_total{id="#anon-parser0",result="discarded"} 0
syslogng_parsed_events_total{id="#anon-parser0",result="processed"} 0
syslogng_scratch_buffers_bytes 0
syslogng_scratch_buffers_total 3
syslogng_tagged_events_total{id=".source.#anon-source0",result="processed"} 0
syslogng_tagged_events_total{id=".source.s_network",result="processed"} 2

$ sbin/syslog-ng-ctl stats prometheus --with-legacy-metrics

syslogng_src_facility_processed{stat_instance="18"} 0
syslogng_dst_network_memory_usage{id="d_dest#0",stat_instance="tcp,127.0.0.1:5555"} 0
syslogng_src_host_processed{id="s_network#1",stat_instance="AnnoW"} 2
syslogng_global_sdata_updates_processed 0
syslogng_global_msg_clones_processed 0
syslogng_global_internal_queue_length_processed 0
syslogng_global_payload_reallocs_processed 4
...

syslog-ng stats labels are similar to the labels in the Prometheus data model:
https://prometheus.io/docs/concepts/data_model/

The (id, labels) pair identifies a new-style key. The order of the labels
should not matter, and a label with an empty value should be equivalent to
a label non-existing, but the current implementation is rudimentary and
does not take these considerations into account.

The old (component, instance , id) keying is marked as legacy; any of the legacy functions should not be used when introducing new metrics.

Questions:

• when should we remove stats-log?

Related PR: #4318

[github-actions]

No news file has been detected. Please write one, if applicable.

[alltilla]

Reviewed until stats: add labeling capability, inclusive. It looks good, I only had UT related comments.

I will check the last 2 commits soon.

[kira-syslogng]

Build FAILURE

[kira-syslogng]

Build FAILURE

[kira-syslogng]

Build FAILURE

[MrAnno]

@kira-syslogng do stresstest

[kira-syslogng]

Kira-stress-test: Build FAILURE

[MrAnno]

@kira-syslogng do stresstest

[MrAnno]

@HofiOne Can you help us fix the stress test?

15:53:49 # An exception occured:
15:53:49 # Type: IOError
15:53:49 # Message: [Errno 28] No space left on device

https://ci.syslog-ng.com/jenkins/job/kira-performance-test-disk-buffer/645/console

[kira-syslogng]

Kira-stress-test: Build FAILURE

[MrAnno]

@kira-syslogng do stresstest

[kira-syslogng]

Kira-stress-test: Build SUCCESS