Clangd consuming a lot of memory once i clicked/hover to definition of function

[learneachday]

Hi,
I am running clangd for my project with remote-index server configured, it is seen that compile-commands.json which is present in project root itself consumed around 1.4GB . The problem here is that when i hover or do Go to definition then memory spikes and keep on increasing it rises to around 32GB . Is this a known problem that memory can rise to such level what can be the workaround for the same. Please suggest. I am using clangd V21 .

Thanks

[HighCommander4]

compile-commands.json which is present in project root itself consumed around 1.4GB

Do you mean the size of compile_commands.json on disk is 1.4 GB?

How many entries does it contain?

[HighCommander4]

What client (editor) are you using? If your client supports clangd's memory usage protocol extension, as e.g. VSCode does (accessed via the "clangd: Show memory usage" command), it would be interesting to use it to understand what is taking up the server's memory.

[learneachday]

I am using vscode, i checked memory usage but its not showing not much memory as can be seen below

There were two files opened which occupied in total of 55MB.

I have checked with copilot and it was pointing me to AST, this is the memory breakdown i got to know.

Can you suggest a way to overcome it or if there is some setting we can use. I tried background-indexing=false but did not work also what i understood from background-indexing->false that it will not index file which is present in remote as I am using project-root pointing to a location where full code is present and remote-index-server to point to the symbol tables and will only be indexing from files which is currently opened.

[HighCommander4]

Ok, so it's something not reflected in the memory usage report that's using a lot of memory. That's unfortunate.

One other idea that comes to mind is to check whether the memory usage is somehow a function of the number of entries in compile_commands.json. You could use a tool like jq to filter the entries in your compile_commands.json to a much smaller subset (say 10% of the original size), and see whether the memory exhaustion still occurs if using a compile_commands.json containing just that subset.

[learneachday]

ok,
I tried your approach of creating a filtered compile commands.json file of 10% of original compile-commands.json while keeping two files entries and all other random entries and i did not faced memory exhaustion issue, it went till 2.2G and then no more exhaust of memory.
I still want to understand what is the best approach here to not exhaust the memory.

[HighCommander4]

The finding may suggest that with clangd's current implementation, it needs more than 32GB of RAM to work on a codebase of this size.

It's possible that there are opportunities to optimize memory usage but they will likely require some investigation / debugging / memory profiling to find.