The Gatekeeper, or a project gating system
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

monitoring.rst 12KB

title

Monitoring

Monitoring

Statsd reporting

Zuul comes with support for the statsd protocol, when enabled and configured (see below), the Zuul scheduler will emit raw metrics to a statsd receiver which let you in turn generate nice graphics.

Configuration

Statsd support uses the statsd python module. Note that support is optional and Zuul will start without the statsd python module present.

Configuration is in the statsd section of zuul.conf.

Metrics

These metrics are emitted by the Zuul scheduler:

zuul.event.<driver>.<type>

Zuul will report counters for each type of event it receives from each of its configured drivers.

zuul.tenant.<tenant>.pipeline

Holds metrics specific to jobs. This hierarchy includes:

<pipeline name>

A set of metrics for each pipeline named as defined in the Zuul config.

all_jobs

Number of jobs triggered by the pipeline.

current_changes

The number of items currently being processed by this pipeline.

project

This hierarchy holds more specific metrics for each project participating in the pipeline.

<canonical_hostname>

The canonical hostname for the triggering project. Embedded . characters will be translated to _.

<project>

The name of the triggering project. Embedded / or . characters will be translated to _.

<branch>

The name of the triggering branch. Embedded / or . characters will be translated to _.

job

Subtree detailing per-project job statistics:

<jobname>

The triggered job name.

<result>

A counter for each type of result (e.g., SUCCESS or FAILURE, ERROR, etc.) for the job. If the result is SUCCESS or FAILURE, Zuul will additionally report the duration of the build as a timer.

current_changes

The number of items of this project currently being processed by this pipeline.

resident_time

A timer metric reporting how long each item for this project has been in the pipeline.

total_changes

The number of changes for this project processed by the pipeline since Zuul started.

resident_time

A timer metric reporting how long each item has been in the pipeline.

total_changes

The number of changes processed by the pipeline since Zuul started.

wait_time

How long each item spent in the pipeline before its first job started.

zuul.executor.<executor>

Holds metrics emitted by individual executors. The <executor> component of the key will be replaced with the hostname of the executor.

merger.<result>

Incremented to represent the status of a Zuul executor's merger operations. <result> can be either SUCCESS or FAILURE. A failed merge operation which would be accounted for as a FAILURE is what ends up being returned by Zuul as a MERGER_FAILURE.

builds

Incremented each time the executor starts a build.

starting_builds

The number of builds starting on this executor and a timer containing how long jobs were in this state. These are builds which have not yet begun their first pre-playbook.

The timer needs special thoughts when interpreting it because it aggregates all jobs. It can be useful when aggregating it over a longer period of time (maybe a day) where fast rising graphs could indicate e.g. IO problems of the machines the executors are running on. But it has to be noted that a rising graph also can indicate a higher usage of complex jobs using more required projects. Also comparing several executors might give insight if the graphs differ a lot from each other. Typically the jobs are equally distributed over all executors (in the same zone when using the zone feature) and as such the starting jobs timers (aggregated over a large enough interval) should not differ much.

running_builds

The number of builds currently running on this executor. This includes starting builds.

paused_builds

The number of currently paused builds on this executor.

phase

Subtree detailing per-phase execution statistics:

<phase>

<phase> represents a phase in the execution of a job. This can be an internal phase (such as setup or cleanup) as well as job phases such as pre, run or post.

<result>

A counter for each type of result. These results do not, by themselves, determine the status of a build but are indicators of the exit status provided by Ansible for the execution of a particular phase.

Example of possible counters for each phase are: RESULT_NORMAL, RESULT_TIMED_OUT, RESULT_UNREACHABLE, RESULT_ABORTED.

load_average

The one-minute load average of this executor, multiplied by 100.

pause

Indicates if the executor is paused. 1 means paused else 0.

pct_used_ram

The used RAM (excluding buffers and cache) on this executor, as a percentage multiplied by 100.

zuul.nodepool.requests

Holds metrics related to Zuul requests and responses from Nodepool.

States are one of:

requested

Node request submitted by Zuul to Nodepool

canceled

Node request was canceled by Zuul

failed

Nodepool failed to fulfill a node request

fulfilled

Nodes were assigned by Nodepool

<state>

Records the elapsed time from request to completion for states failed and fulfilled. For example, zuul.nodepool.request.fulfilled.mean will give the average time for all fulfilled requests within each statsd flush interval.

A lower value for fulfilled requests is better. Ideally, there will be no failed requests.

<state>.total

Incremented when nodes are assigned or removed as described in the states above.

<state>.size.<size>

Increments for the node count of each request. For example, a request for 3 nodes would use the key zuul.nodepool.requests.requested.size.3; fulfillment of 3 node requests can be tracked with zuul.nodepool.requests.fulfilled.size.3.

The timer is implemented for fulfilled and failed requests. For example, the timer zuul.nodepool.requests.failed.size.3.mean gives the average time of 3-node failed requests within the statsd flush interval. A lower value for fulfilled requests is better. Ideally, there will be no failed requests.

<state>.label.<label>

Increments for the label of each request. For example, requests for centos7 nodes could be tracked with zuul.nodepool.requests.requested.centos7.

The timer is implemented for fulfilled and failed requests. For example, the timer zuul.nodepool.requests.fulfilled.label.centos7.mean gives the average time of centos7 fulfilled requests within the statsd flush interval. A lower value for fulfilled requests is better. Ideally, there will be no failed requests.

current_requests

The number of outstanding nodepool requests from Zuul. Ideally this will be at zero, meaning all requests are fulfilled. Persistently high values indicate more testing node resources would be helpful.

zuul.mergers

Holds metrics related to Zuul mergers.

online

The number of Zuul merger processes online.

jobs_running

The number of merge jobs running.

jobs_queued

The number of merge jobs waiting for a merger. This should ideally be zero; persistent higher values indicate more merger resources would be useful.

zuul.executors

Holds metrics related to Zuul executors.

online

The number of Zuul executor processes online.

accepting

The number of Zuul executor processes accepting new jobs.

jobs_running

The number of executor jobs running.

jobs_queued

The number of jobs allocated nodes, but queued waiting for an executor to run on. This should ideally be at zero; persistent higher values indicate more executor resources would be useful.

zuul.geard

Gearman job distribution statistics. Gearman jobs encompass the wide variety of distributed jobs running within the scheduler and across mergers and executors. These stats are emitted by the gear library.

running

Jobs that Gearman has actively running. The longest running jobs will usually relate to active job execution so you would expect this to have a lower bound around there. Note this may be lower than active nodes, as a multiple-node job will only have one active Gearman job.

waiting

Jobs waiting in the gearman queue. This would be expected to be around zero; note that this is not related to the backlogged queue of jobs waiting for a node allocation (node allocations are via Zookeeper). If this is unexpectedly high, see debug_gearman for queue debugging tips to find out which particular function calls are waiting.

total

The sum of the running and waiting jobs.

As an example, given a job named myjob in mytenant triggered by a change to myproject on the master branch in the gate pipeline which took 40 seconds to build, the Zuul scheduler will emit the following statsd events:

  • zuul.tenant.mytenant.pipeline.gate.project.example_com.myproject.master.job.myjob.SUCCESS +1
  • zuul.tenant.mytenant.pipeline.gate.project.example_com.myproject.master.job.myjob.SUCCESS 40 seconds
  • zuul.tenant.mytenant.pipeline.gate.all_jobs +1