fix: Spawn task for block proof verification (#288) #308
Conversation
Verifying a block proof takes on the order of hundreds of microseconds. It is good practice to wrap such calls inside of a `spawn_blocking` invocation, which is what this commit does. Besides observing good practice, it seems as though this change fixes issue #282. The reported crash was not yet observed with this change in place. However, we do not understand the relevant mechanics or have an educated guess as to why this change causes the issue to go away. For all we know, the issue might appear again even with this change. In order to use `spawn_blocking`, the function in question must be made `async`. This change percolates up the call graph. The call to `triton_vm::verify` that pinpoints issue #282, or more precisely, the `spawn_blocking` wrapper that contains it, is sandwiched in between log statements. So if the node hangs again, the node operator will be able to pinpoint the cause.
|
Based on this thought from the discussion in #282 --
-- I tend to think that this change limits the crash to the thread/task as opposed to the whole application, but does not fix it. While I do think it is good practice to wrap long-running computations like |
|
With #315 I think issue #282 is comprehensively resolved and so we can proceed with this PR. Any objections, @dan-da, @Sword-Smith? |
triton_vm::verifyinspawn_blocking.asyncand percolateasyncmarkers.All observations point to this change eliminating issue #282. However, we do not have a mechanical explanation as to what causes the apparent fix.