Monitoring: Add a Prometheus-compatible API endpoint

[lastzero]

As an admin, I want to monitor the health and performance of PhotoPrism using a common tool like Prometheus.

This PR by @brandon1024 added a Prometheus-compatible GET /api/v1/metrics endpoint to our API:

• API: Expose Prometheus-style metrics endpoint #3730

See comments below for further details.

[RichiH]

You might call me biased, but I would prefer Prometheus.

Influx can understand Prometheus data format and by using client_golang, you would get the upgrade path to OpenMetrics for free.

[lastzero]

What could be advantages / disadvantages for our specific application? What metrics would you collect?

[RichiH]

It is very unlikely that you would be unhappy with using Prometheus client libraries. You get baseline process metrics automagically, so I would suggest starting with vanity/boundary metrics (amount of pictures, storage size), usage (how often are certain endpoints hit, how long do they take to load, etc), and "put a counter everywhere you would be tempted to put a debug statement". That should immediately be useful and also allow you to get a feeling of what you find useful, or fun. Sent by mobile; please excuse my brevity.

…

On Thu, Jan 23, 2020, 22:40 Michael Mayer ***@***.***> wrote: What could be advantages / disadvantages for our specific application? What metrics would you collect? — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#213?email_source=notifications&email_token=AAFYII7Z4ZFW3RLYQ6JDERTQ7IFDPA5CNFSM4KKDJSOKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEJY6KUA#issuecomment-577889616>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAFYIIZ3FMWUFZSDLPA5AQ3Q7IFDPANCNFSM4KKDJSOA> .

[lastzero]

Compared both recently and found that InfluxDB was developed after Prometheus to fix shortcomings for certain use cases. AFAIK InfluxDB needs less storage and can have higher performance, but that really doesn't matter for a private user. Question was meant to learn, not to suggest users would be unhappy with Prometheus.

[RichiH]

Ah, OK. InfluxDB predates Prometheus by quite a bit. Using Prometheus client allows you to use both Prometheus and InfluxDB, and others, as they added support for Prometheus. Performance is roughly the same, but I think we are slightly ahead of Influx in storage and performance. Anything else you want to know about metrics etc, just ask :) Sent by mobile; please excuse my brevity.

…

On Sun, Jan 26, 2020, 11:59 Michael Mayer ***@***.***> wrote: Compared both recently and found that InfluxDB was developed after Prometheus to fix shortcomings for certain use cases. AFAIK InfluxDB needs less storage and can have higher performance, but that really doesn't matter for a private user. Question was meant to learn, not to suggest users would be unhappy with Prometheus. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#213?email_source=notifications&email_token=AAFYII64ORBMK7QSDAQMSBTQ7VUHXA5CNFSM4KKDJSOKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEJ5RB4Q#issuecomment-578490610>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAFYII5QDEAWIQOAZTKNSJ3Q7VUHXANCNFSM4KKDJSOA> .

[haozhou]

Just curious which KPI you'd like to take for health monitoring?
Disk usage or CPU usage is application-independent

[mmontes11]

Any update about this? It would be a huge plus! 📈🚀

[lastzero]

Nope. Working on face detection right now. You can do monitoring via Traefik already if you just want to see requests.

[centralhardware]

Just curious which KPI you'd like to take for health monitoring? Disk usage or CPU usage is application-independent

hm, looks like all we need it's node exporter for node monitoring and mysql_exporter for database Monitoring.
Maybe we can monitor count of recognized faces or count of photos but i don't think it's useful enough

[RichiH]

For direct instrumentation, it would be more common to expose e.g. latency buckets for processing time of photos or requests. For data that's in MySQL, or on the local machine, mysql_exporter & node_exporter are better, agreed.

[MostFrabjous]

I'd vote for some API endpoints that expose statistics about the photoprism instance. Like

• number of originals|hidden|error|photos|places|albums|people|[new] faces|etc.
• sync status? (I'm not using sync)