I'm still working through the best schema, but it may be useful to have the ability to add a prefix to all measurements that get sent to InfluxDb. For example, we send measurements with names like "cron" and "changelog". In InfluxDb measurements may come from other source, though, such as telegraf. To reduce the risk of collision it may be useful to have the ability to prefix all measurements (e.g. magento_cron, magento_chagenlog, etc...)

JS to post to a Magento route

• route
• URL
• JS http request count
• JS http request bytes downloaded
• CSS http request count
• CSS http request bytes downloaded
• IMG http requests count
• IMG http requests bytes donwloaded
• Page timings

Magento route stores it in Redis where it is queued to be synced to InfluxDb by shell/sd_influxdb.php

Per #26 (comment) sprintf is cleaner. Need to review all the measurements.

Currently groups must be configured via XML. It could be useful to have the ability to do this with a GUI.

We may want to vary the retention policy by measurement rather than just always writing to the DEFAULT RP. This could potentially be handled in the config. E.g.

<sd_influxdb>
    <groups>
        <one_minute>
            <inventory>
                <model>sd_influxdb/measurement_inventory</model>
                <retention_policy>ninety_days</retention_policy>
            </inventory>
        </one_minute>
    </groups>
</sd_influxdb>

It would be cool if Telegraf supported running specific Redis queries, but I don't think they do.

What happens if InfluxDb goes down

• In progress requests pile up and are never deleted
• If this isn't addressed, redis store could fill up which could create performance issues for the application as every request will try to write to the redis store.

Solutions

• Set an expiry when creating InProgress keys. We lose the data if InfluxDb goes down for an extended period of time, but we decrease possibility of jeopardizing application performance
• Also, check for InProgress keys during send. Use timestamp added to in progress flag when sending to InfluxDb

E.g.

X-SD-InfluxDb-Route catalog/category/view

We can then record that to a custom log and stream it to InfluxDb with https://github.com/influxdata/telegraf/tree/master/plugins/inputs/logparser

Thus we can calculate increase in 5XX errors on a per route basis (e.g. Increase in 5XX to /checkout/cart/add might go undetected when not looking on a per route basis).

Would be interesting to track the highest version Id in each changelog table as well as the current version in the metadata table. We could learn about whether lag is an issue, understand cache invalidations due to partial reindexes and, I'm sure, more...

Per #26 (comment)

There is one the changelog measurement, obviously, but worth checking the other measurements if they can be cleaned up similarly.