HibiAPI
HibiAPI copied to clipboard
mixmoe
Too much memory use & possible memory leak
This is Nanahira, the maintainer of MyCard, also the instance owner of https://hibi.moecube.com. Recently we saw the instance ate up 4 GB of our server memory, blew up our monitor system's alarm. Is this software consuming that amount of memory, or perhaps my own misconfiguration? Looking for help, thanks.
Hello Nanahira, thank you for maintaining the instance and for being the owner of hibi.moecube.com.
Regarding the issue of the instance consuming 4 GB of server memory and triggering the monitoring system's alarm, it may be caused by caching enabled by default. It is normal for this to occur when the request volume is high. You can refer to #114 for more information. I recommend upgrading to the latest version of HibiAPI, including all dependencies, as this may help with the memory usage issue.
However, if memory still grows with low request volume, there may be a potential memory leak in the program that requires further analysis.
With a careful check of the logs of both our gateways and CDN, we found there were 1000 requests per day, which isn't a high number. Is that amount of memory and request rate looking fine on other instances, especially the official one? In addition, we would try to migrate to Redis caching, trying to solve this problem. Thanks for your reply anyway.
For your reference, this is the memory graph of the public demo api.obfs.dev, it uses disk cache as cache backend and holds average 120–160 requests per minute.
It does appear that even with the use of diskcache, the memory usage is still slowly increasing. Given the reported request volume, it is likely that there is a memory leak present in the program. Debugging this issue may not be straightforward, and the rate at which the memory usage is increasing is relatively slow. I will try to add an APM to monitor the detailed memory profile of the program, and then work to fix the issue (I doubt that the issue may be caused by an upstream library rather than the code in the repository, which could make it more difficult to fix). So leave this issue open.
As a workaround, you can set up a scheduled restart to reduce the impact of the memory leak.
If you're using the default config then you have to set how much in-memory it can use since on default the size is not defined so it ends up using all your memory. For example I use mem://?size=256 which works fine for low traffic
