This RFC is to introduce a new metrics source to consume metrics from the
Apache HTTP Server (httpd). The high level plan is
to implement a scrapper similar to the existing prometheus
source that will scrape
the Apache HTTP Server stats endpoint (provided by
mod_status
) on an
interval and publish metrics to the defined pipeline.
This RFC will cover:
- A new source for Apache metrics
This RFC will not cover:
- Generating metrics from Apache logs
Users running httpd want to collect, transform, and forward metrics to better observe how their webservers are performing.
I expect to largely copy the existing prometheus source and modify it to parse the output of the httpd status page which looks like:
localhost
ServerVersion: Apache/2.4.46 (Unix)
ServerMPM: event
Server Built: Aug 5 2020 23:20:17
CurrentTime: Friday, 21-Aug-2020 18:41:34 UTC
RestartTime: Friday, 21-Aug-2020 18:41:08 UTC
ParentServerConfigGeneration: 1
ParentServerMPMGeneration: 0
ServerUptimeSeconds: 26
ServerUptime: 26 seconds
Load1: 0.00
Load5: 0.03
Load15: 0.03
Total Accesses: 30
Total kBytes: 217
Total Duration: 11
CPUUser: .2
CPUSystem: .02
CPUChildrenUser: 0
CPUChildrenSystem: 0
CPULoad: .846154
Uptime: 26
ReqPerSec: 1.15385
BytesPerSec: 8546.46
BytesPerReq: 7406.93
DurationPerReq: .366667
BusyWorkers: 1
IdleWorkers: 74
Processes: 3
Stopping: 0
BusyWorkers: 1
IdleWorkers: 74
ConnsTotal: 1
ConnsAsyncWriting: 0
ConnsAsyncKeepAlive: 0
ConnsAsyncClosing: 0
Scoreboard: ________________________________________________________W__________________.....................................................................................................................................................................................................................................................................................................................................
I'll use this to generate the following metrics:
apache_up
(gauge)apache_uptime_seconds_total
(counter)apache_accesses_total
(counter; extended)apache_sent_kilobytes_total
(counter; extended)apache_duration_seconds_total
(counter; extended)apache_cpu_seconds_total{type=(system|user|cpu_children_user|cpu_children_system)}
(gauge; extended)apache_cpu_load
(gauge; extended)apache_workers{state=(busy|idle)}
(gauge)apache_connections{state=(closing|keepalive|writing|total)}
(gauge)apache_scoreboard_waiting{state=(waiting|starting|reading|sending|keepalive|dnslookup|closing|logging|finishing|idle_cleanup|open}
(gauge)
Metrics labeled extended
are only available if ExtendedStatus
is enabled
for Apache. This is the default in newer versions (>= 2.4; released 2012), but
purportedly increases CPU
load
so some users may turn it off. If it is off, they simply won't have those
metrics published.
I figure we probably don't want metrics for:
- System Load (should be handled by a
cpu
or similar metrics source)
Metrics will be labeled with:
endpoint
the full endpoint (sans any basic auth credentials)host
the hostname and port portions of the endpoint
Users will be instructed to setup
mod_status
and
enable
ExtendedStatus
.
The following additional source configuration will be added:
[sources.my_source_id]
type = "apache_metrics" # required
endpoints = ["http://localhost/server-status?auto"] # required, default
scrape_interval_secs = 15 # optional, default, seconds
namespace = "apache" # optional, default, namespace to put metrics under
Some possible configuration improvements we could add in the future would be:
response_timeout
; to cap request lengthstls
: settings to allow setting specific chains of trust and client certsbasic_auth
: to set username/password for use with HTTP basic auth; we'll allow this to be set in the URL too which will work for now
But I chose to leave those out for now given the Prometheus source doesn't support them either. We could add support to both at the same time (see Future Work section below).
Datadog's plugin has numerous more options we could also consider in the future.
The host
key will be set to the host parsed out of the endpoint
.
Apache HTTP Server is a fairly common webserver. If we do not support ingesting metrics from it, it is likely to push people to use another tool to forward metrics from httpd to the desired sink.
As part of Vector's vision to be the "one tool" for ingesting and shipping observability data, it makes sense to add as many sources as possible to reduce the likelihood that a user will not be able to ingest metrics from their tools.
- Additional maintenance and integration testing burden of a new source
We could not add the source directly to Vector and instead instruct users to run Telegraf and point Vector at the exposed Prometheus scrape endpoint. This would leverage the already supported telegraf Apache input plugin
I decided against this as it would be in contrast with one of the listed principles of Vector:
One Tool. All Data. - One simple tool gets your logs, metrics, and traces (coming soon) from A to B.
On the same page, it is mentioned that Vector should be a replacement for Telegraf.
You SHOULD use Vector to replace Logstash, Fluent*, Telegraf, Beats, or similar tools.
If users are already running telegraf though, they could opt for this path.
We could model this as a generic http_scrape
source that has a type
(or
codec
?) that would determine how what type of endpoint it is scraping.
I would err away from this as I think there is some risk that some "HTTP scrape" endpoints will need source specific configuration. We could do this later if this does not end up being the case and they really are all the same.
One downside of this is that I think it'd be less discoverable than a first-class source for each type of endpoint we support scraping.
- Do we want to apply any metric labels based on the other information
available via the status page? I could see labeling the
url
at least. Answer: label withhost
andendpoint
as described above. - Do we want to have one apache_metrics source able to scrape multiple endpoints? Answer: yes, the config has been updated to allow multiple endpoints.
- Are there preferences between
apache
orhttpd
for the nomenclature? I feel likeapache
is more well-known thoughhttpd
is more accurate. Answer: standardize onapache
. - Should the
host
key include the port from theendpoint
, if any? Or just the hostname. Answer: include the port.
Incremental steps that execute this change. Generally this is in the form of:
- Submit a PR with the initial sink implementation
I think one thing that would make sense would be to refactor the sources based
on HTTP scraping to share a base similar to how our sinks that rely on http
are factored (splunk_hec
, http
, loki
, etc.). This allows them to share
common configuration options for their behavior.
My recommendation is to implement this and the
nginx
metrics source and
then figure out where the seams our to pull out an HttpScrapeSource
module
that could be used by this source, the nginx
source, and the prometheus
source.