For clusters.processes.[instance].disk, there is free_bytes as well as total_bytes (both are given as quantities of RAM, while the key name is ‘disk’). I assume it is indeed RAM information and “disk” label is for symmetry/historic
I noticed, with the comments, it isn’t really a valid json. Is there a way for me to submit corrections? pull request on the github repo? For example, a comma is needed after value for key “database_lock_state”.
My db (“6.3.15”) returns a “logs” key under “clusters”. The document above has only “old_logs”. Is this the way keys/metrics are deprecated? Any way to get a version of this documentation that matches the released FDB?
Among OSS projects with prometheus support, this one
seems most comprehensive, although changes in last 2 years would be missing.
I also appreciate any other pointers to libraries that deal with status json output.
I do understand that status json doesn’t have any preset schema and all keys can change anytime.
I see you stambled across my Prometheus exporter I wrote this in order to quickly export some metrics for some homemade chaos-monkey experiment. I’m not using it, and I’m pretty sure it is not safe (safe as I’m sure the model is correct and it can deserialize properly across multiple fdb versions).
I will install a fdb cluster in the short future and I may rewrite it to make it more useful. In the meantime, you can have a look at this discussion where you can see how others are monitoring and using the status json.
I guess you can still build the relevant documentation for a tag/branch by running ninja docpreview.
Free and total bytes are both of the disk, and the comment in our documentation mentioning RAM is incorrect. In addition to these, in the storage and log roles there are some metrics that include space details. These include *_available_bytes (e.g. kvstore_available_bytes), which measures how much space is usable on disk and is in most cases a more interesting number than free_bytes.
Yes, if you would like to submit corrections to the website documentation, feel free to add a pull request. You are correct that the website docs are not valid JSON because of the comments, but besides that I think the intent is that they should be valid. There is also a second file which is meant to be kept in sync with the website docs and which I think should be valid (though I haven’t confirmed this). The two files are:
old_logs here refers to the old generation of transaction logs after a recovery occurs and we recruit new ones and is not meant to imply deprecation. It seems that 6.3 changed what we report here and renamed this field to logs, but that is not reflected in the documentation. I’ll see if I can have that updated.
We(Clever Cloud) are investing in FDB, but we haven’t deployed a cluster yet. We should start this somewhere around September/October. We have some PoC of tools to gain observability over a FDB cluster, including:
a new Prometheus exporter. This will be fairly similar to the one written in go, but it will be:
not tied to a fdb version,
capable of generating metrics directly by parsing the json dynamically while recognize some patterns like machines/processes key as tags, and enum/messages handling
written in rust.
a tracing exporter. This is based on the 7.0 release which has some Tracing features thanks to @markus.pilman’s work. Parsing and sending to Jaeger is done, I just need to figure out how to handle the Multiple parent spans in Jaeger.
Both of them will be open-sourced, once we have them battle-tested on our side.