Confirm if there is memory leak

[mhchia]

What's wrong?

In #6 (comment), we encountered OOM and suspected there was memory leak in the latest multi-threaded prover. It was worked around by increasing wasm memory to 4 GB.

Possible Cause of OOM

The single-threaded prover seems to have a similar memory usage as the multi-threaded prover, but it doesn't encounter an OOM. Thus, I suspect OOM might not result from a memory leak. It happens in multi-threaded prover just because we enable the rustflags atomics (required by wasm-bindgen-rayon) and max-memory is automatically set to 1 GB.

Actions

• Confirm single-threaded and multi-threaded prover max-memory without setting it explicitly. If they are different, it's possible that OOM is just because multi-thread has a 1 GB max memory while single-threaded doesn't, and thus OOM is possibly not because of the memory leak

Resources

• rustc linker adds max-memory=1GB when atomics is enabled

[mhchia]

Using wasm-objdump, we can see there is a max set when using a multi-threaded prover. However, there isn't a maximum memory set when using a single-threaded prover. This could explain why we didn't encounter OOM in a single-threaded prover even if it uses more than 1 GB memory.

Multi-threaded (nightly + rustflags atomics)

max is set to 1 GB (1 GB = 16384 * 64 KB, 16384 pages and 64 KB per page).

$ wasm-objdump --details --section=import target/wasm32-unknown-unknown/release/tlsn_extension_rs.wasm| grep memory
- memory[0] pages: initial=96 max=16384 shared <- env.memory

Single-threaded (stable)

No max is set

$ wasm-objdump --details --section=import target/wasm32-unknown-unknown/release/tlsn_extension_rs.wasm| grep memory
...

[mhchia]

There seems no obvious evidence that a memory leak is occurring now. Close it for now and we can reopen it if it's still relevant