gh-117645: Increase WASI stack size from 512 KiB to 8 MiB

[vstinner]

Increase also the initial memory from 10 MiB to 20 MiB.

Reenable test_dynamic on WASI release build.

• Issue: test_dynamic fails on wasm32 WASI 8Core 3.x buildbot #117645

[vstinner]

!buildbot wasi

[bedevere-bot]

🤖 New build scheduled with the buildbot fleet by @vstinner for commit 2736f25 🤖

The command will test the builders whose names match following regular expression: wasi

The builders matched are:

• wasm32-wasi PR
• wasm32-wasi Non-Debug PR
• wasm32 WASI 8Core PR

[vstinner]

The 8 MiB limit (8 388 608 bytes) is already used by the wasmtime command:

test.pythoninfo:

sysconfig[HOSTRUNNER]: wasmtime run --wasm max-wasm-stack=8388608 --wasi preview2 --env PYTHONPATH=/$(shell realpath --relative-to /opt/buildbot/kushaldas-wasm/3.x.kushaldas-wasi.wasi.nondebug/build/build_oot/host/../.. /opt/buildbot/kushaldas-wasm/3.x.kushaldas-wasi.wasi.nondebug/build/build_oot/host)/$(shell cat pybuilddir.txt):/Lib --dir ../..::/

The value is set at:

$ grep 8388608 Tools/wasm/ -R
Tools/wasm/wasi.py: # The 8388608 value comes from `ulimit -s` under Linux
Tools/wasm/wasi.py: "--wasm max-wasm-stack=8388608 "
Tools/wasm/wasm_build.py: "--wasm max-wasm-stack=8388608 "

I'm not sure of the difference between the linker -z stack-size=8388608 option and wasmtime option max-wasm-stack=8388608. According to the issue, the stack doesn't seem to grow from 512 KiB to 8 MiB automatically: #117645 (comment)

With this change, test_dynamic works as expected (on a release build) with 8 MiB stack.

[vstinner]

@brettcannon: Would you mind to review this change? Are you ok to increase the initial stack from 512 KiB to 8 MiB?

[vstinner]

test_dynamic pass again on "buildbot/wasm32 WASI 8Core PR" (built in release mode). Previously, test_dynamic was failing on this buildbot.

[vstinner]

Note: I also tried to leave the initial stack unchanged (512 KiB), and only increase the memory size (to 20 MiB): -z stack-size=524288 -Wl,--stack-first -Wl,--initial-memory=20971520. test_dynamic fails in this case (release build). We need to increase the stack size in the linker if we want test_dynamic to pass: what this PR does.

[vstinner]

I'm not convinced that wasmtime can grow the stack and/or the total memory size. So I changed the configure.ac comment.

[encukou]

I think this definitely needs a WASI expert. It looks a bit like tuning generic build settings for the specific buildbot.
@ericsnowcurrently, do you want to take a look?

[vstinner]

I did another test:

• 512 KiB initial stack
• 40 MiB initial memory

test_dynamic fails with:

 2: memory fault at wasm address 0x1000005b0 in linear memory of size 0x2800000
3: wasm trap: out of bounds memory access

It confirms that the wasmtimecannot grow the stack: we have to set it to 8 MiB directly.

But it can grow the "total" memory (address space). Example on a build with 40 MiB initial memory:

# cross-build/wasm32-wasi/python.sh Python 3.13.0a5+ (heads/wasi_stack-dirty:a69ffa75eb, Apr 15 2024, 13:36:08) [Clang 16.0.0 ] on wasi
Type "help", "copyright", "credits" or "license" for more information.
>>> x=b'x' * (1024 * 1024 * 60)
>>> len(x)
62914560

I can allocate 60 MiB without any issue.

[vstinner]

It seems like nobody is available to review this change. I plan to merge my PR at the end of the week, unless someone comes and wants to review it :-)

The change reenables a test_dynamic test which is currently always skipped on WASI. IMO it will unlock other use cases which are currently limited by the maximum stack memory.

IMO a stack of 512 KiB is too small for Python, and sadly --wasm max-wasm-stack=8388608 option to wasmtime doesn't work: it fails to grow the stack.

[ericsnowcurrently]

LGTM

Thanks for doing this! I've left one comment, but I'll leave it to you discretion.

[vstinner]

Merged, thanks for the review @ericsnowcurrently.