Memory measurement and limits for Squirrel #7516
Conversation
nielsmh
changed the title
|
I think this is functionally complete now. Two remaining questions:
|
|
As for the memory limit, I think that this memory limit should default depending on your system's RAM. I also propose that the RAM limit for AI/GS be configurable (both total and per-script, hidden from Advanced Settings), accepting k/K, m/M, g/G for 1024 (1 KiB), 1048576 (1 MiB), and 1073741824 (1 GiB) respectively. The reason for that is that some people may have computers with 16GB RAM plus, and want to have their AI/GS be able to utilize more of that, while other computers may only have 512 MB to 1 GB RAM and/or don't want their AI/GS taking up a lot of RAM to prevent out-of-memory errors. Edit: Commit f55b811 has made the max memory limit per script configurable between 8MB (required for regression/dummy AI) and 8GB (1/2-1/4 of the average high-end gaming PC's RAM) |
|
Good luck finding a reliable cross-platform way to determine amount of system RAM, let alone free RAM. |
nielsmh
force-pushed
the
sqmem
branch
2 times, most recently
from
e86c341 to
9a5d17d
Compare
|
Resolved conflict with saveload version. |
|
This PR needs to be rebased |
This can avoid out-of-memory situations due to single scripts using up the entire address space. Instead, scripts that go above the maximum are killed. The maximum is default 1 GB per script, but can be configured by a setting.
Memory management for AI and GS is bad, everything comes out of the runtime library's default heap, so a single script running loose can exhaust all memory available to the OpenTTD process and cause a hard crash of the game.
This PR extends Squirrel to support custom a custom allocator (which has to be "context switched" manually), and so far uses it to keep track of allocation size for each running script. The end goal of this PR is to implement an arena allocator that will allow putting hard limits on how much memory a single script can use.
Obligatory screenshot:
Inspired by report in #7513 but it turns out the problem there may not be related to OoM conditions.