Following from our previous report comparing CrateDB against MongoDB, we can now also present TSBS benchmark results for InfluxDB.
As with the previous report, CrateDB sponsored this work, but did not participate in the benchmarking and did not have editorial control over the report. (In fact, several suggestions from the review phase were rejected in order to keep the report fair and objective.) As the recipient of the work, they did however have the right to decide whether to publish the results or not. They chose to publish.
The configuration is exactly the same as in the previous report – minus disk and IOPS capacity that was vastly over provisioned in the previous report. CPU remains the bottleneck also in these tests, the IOPS is never near saturation.
The DSI framework is starting to show its power. Thanks to the generalization work done for testing CrateDB, getting InfluxDB deployment automated was now a matter of one or two days…
Except… Authorization! It’s ridiculous, but for all databases I’ve tested recently, figuring out how to login to the damn thing is more time consuming than everything else combined. This is even true for MySQL/MariaDB, which I know well and have used since the previous century! The only database that was easy to connect to is MongoDB, because it is still easy to start it without authentication turned on at all! Unfortunately, if you do that, and leave the port open in the firewall, your data will be deleted faster than you can finish loading it into the database. Which brings you back to needing to figure out how to enable authentication, and now you already lost a couple of days to the ransomware kids.
(Btw, those hackers are doing a great service to the MongoDB community, if you ask me. They are so aggressive, it’s impossible to ever upload any valuable data before they already find you.)
For a backup and restore service 350EUR is actually quite a good price! Pentesting service for free?
But I digress… How about InfluxDB? InfluxDB was by far the most difficult to connect to. Partly this was because I wasted days when I refused to accept that by design a human must open a web browser and copy paste a token. Silly me, should have accepted faster that it’s 2024 and I can’t automate database deployments. (More seriously, is this a strategy to drive production databases to they cloud offering? The self managed version is intended for developers laptops? Makes sense if this is intentional. But we continued testing InfluxDB OSS version 2.7 since there are still users out there intending to use it in production.
Once we got it running, results were interesting. Participating with an LSM-like storage engine, I expected InfluxDB to have really fast writes and less so with reads. But it turns out it was rather opposite, as it is not really an LSM engine, just “like”. Results were also more even as both databases were better in some aspect of TSBS. (See: It’s not always the vendor that commissioned the report that wins everything!)
