| Table of Contents |
|---|
Overview
There are several parallel efforts to capture information about Clowder metrics:
...
The goal is to minimize number of moving parts to capture and store this data. Below is summary of our discussion from 12/7.
Recon / Planning Phase
RabbitMQ Queue & Flask API
...
- Allow for configuration of Amplitude API key (no change)
- If configured, tracking snippet added to every view (no change)
- For the tracked events above, call the new backend SinkService, which will check for configured integrations with Amplitude/Google Analytics/etc and delegate appropriately:
- This will be a new piece of code that will submit to our special RabbitMQ queue/exchange
- If Amplitude is configured, also send to Amplitude via the REST API
- If GA is configured, also send to Google Analytics (NOTE: may be difficult or impossible - see https://stackoverflow.com/questions/15530487/restful-api-and-google-analytics)
- Bonus points: add a backend action that automatically tracks API calls and sends to the SinkService
Legend
- Orange: existing technology
- Blue: likely a new piece that needs to be written
Clowder health monitor(s)
...
Let's consider some different types of events. Assume user and timestamp for all data captured too.
| component | event type | data captured | notes |
|---|---|---|---|
| storage | file uploaded file deleted | fileid, datasetid, spaceid, bytes | |
| extractions | extraction event | message, type (queued or working) | do we care about data traffic downloaded to the extractor containers? |
| traffic | page views resource downloads | url, resourceid bytes | do we care about every page view? this is currently tracking which resources are being viewed but without the full url |
| health | ping update | response time, queue length, other? | |
The Full Picture
Legend
- Orange: existing technology
- Blue: likely a new piece that needs to be written
Refactoring RabbitMQ
Clowder currently has RabbitMqPlugin.scala which contains logic for submitting files and datasets to the appropriate queues in RabbitMQ for scheduling extractions. A client (e.g. pyClowder) can then subscribe to the queue(s) to systematically work through related extraction jobs in a scalable fashion.
We want to preserve the above relationship, while making the underlying code slightly more generalized such that we can use it send an arbitrary message to an arbitrary queue in the configured RabbitMQ instance.
This new generalized service (name TBD, but for now I will call this RabbitMqService.scala) can be used by the existing RabbitMqPlugin.scala (or optionally a simplified rewrite) to submit jobs to extractors in the same way. The new RabbitMqService.scala can also be utilized by new code to send events to the event sink system.
The Event Sink System
As described above, the "event sink" is simply a special exchange in RabbitMQ that is configured to fanout to multiple queues. This way we can siphon different types of events into specific queues based on the intended target.
The plan is to create a set of very thin AMQP event worker clients, which will start up and subscribe to the queue(s) to systematically work through delivering the metrics/event payloads to interested parties.
Emergency Flask API
In order to track events surrounding Clowder downtime, we should offer an alternative/emergency API (via Flask, or similar). This way if Clowder is down or has an outage, we can still continue to collect metrics data by using the exposed Flask API endpoint(s) to submit to the event sink exchange in RabbitMQ, as described above.
The Flask API only needs to be instrumented to collect/submit events/metrics related to Clowder uptime/downtime, maintenance, and outages.
The Flask API only needs to submit directly to RabbitMQ, and does not need to integrate directly with Clowder itself.
Event Workers
MongoDB Worker
A debug worker that will simply echo the given message into a new collection in MongoDB.
This can be used for debugging purposes and to provide a simple pattern/template for creating future event workers.
InfluxDB Worker
Our centralized receiver that will receive all events of all types and parse them into InfluxDB.
The ultimate goal is to use Grafana/InfluxDB as a centralized place from which to view/store all metrics/event data. Grafana then allows us to build graphs, dashboards, and alerts based off of these metrics
Experimental Workers
The tricky part here is that typically these types of analytics/clickstream services tend to integrate directly with the user's browser via JavaScript, and not via a backend service. We have no guarantee that these proposed workers will function as described, but we are hopeful that the resources offered by the associated SDKs will allow for them to fit into our solution.
Amplitude Worker
Amplitude supposedly offers a REST API for submitting events directly from the server-side:
If possible, this will send UI-driven or event-related metrics to Amplitude to track user analytics:
- Demographic (e.g. Age/Location/Language)
- Device Info (e.g. Hardware/OS/Browser)
- User Interactions (e.g. Page Views/Events)
Google Analytics Worker
Google Analytics supposedly once offered an API for submitting events directly from the server-side:
Unfortunately, briefly attempting to follow the links above yielded some 404 errors, so some exploratory work will be necessary here.
If possible, this will send UI-driven or event-related metrics to Google Analytics to track user analytics:
- Demographic (e.g. Age/Location/Language)
- Device Info (e.g. Hardware/OS/Browser)
- User Interactions (e.g. Page Views/Events)