handle deadlocks/blocked tasks in parallel reconstruction #7800
Conversation
beacon_chain/nimbus_beacon_node.nim
Outdated
| if true: | ||
| return | ||
| # # Currently, this logic is broken | ||
| # if true: |
There was a problem hiding this comment.
Can just remove this if entirely, rather than commenting it out.
| localIndices[j] = idxArr[j] | ||
| localCells[j] = cellsArr[j] | ||
| # use the task wrapper which maps string errors to void | ||
| recoverCellsAndKzgProofsTask(localIndices, localCells) |
There was a problem hiding this comment.
there is no need for this seq as far as I can tell - just change the argument type of ...Task to openArray, then use idxArr.toOpenArray(0, columnCount - 1)
|
|
||
| var | ||
| pendingFuts: seq[Flowvar[Result[CellsAndProofs, void]]] | ||
| pendingFuts: seq[Flowvar[Result[CellsAndProofs, void]]] = @[] |
There was a problem hiding this comment.
The = @[] doesn't do anything here.
| pendingFuts: seq[Flowvar[Result[CellsAndProofs, void]]] = @[] | ||
| # Store actual data sequences instead of C pointers | ||
| pendingIndices: seq[seq[CellIndex]] = @[] | ||
| pendingCells: seq[seq[Cell]] = @[] |
There was a problem hiding this comment.
Likewise either of these places. https://nim-lang.org/docs/manual.html#statements-and-expressions-var-statement
Variables are always initialized with a default value if there is no initializing expression. The default
value depends on the type and is always a zero in binary.
where for a sequence, this default is documented to be @[].
| # pre-size sequences so we can index-assign without reallocs | ||
| pendingFuts.setLen(blobCount) | ||
| pendingIndices.setLen(blobCount) | ||
| pendingCells.setLen(blobCount) |
There was a problem hiding this comment.
If these are going to be setLen'd immediately, why not newSeq[Flowvar[Result[CellsAndProofs, void]]](blobCount) (respectively, each seq type) them to begin with?
|
|
||
| for i in 0 ..< dataColumns.len: | ||
| cellIndices[i] = dataColumns[i][].index | ||
| indices[i] = dataColumns[i][].index |
There was a problem hiding this comment.
The redundant/indirect assignment/copying via indices and cells is fairly expensive. Can directly ensure that pendingIndices[spawned] and pendingCells[spawned] are the right side, the have this loop assign into pendingIndices[spawned] item by item directly (respectively, pendingCells[spawned])
Can also use addr pendingIndices[spawned]/addr pendingCells[spawned] can avoid dataColumns.len lookups/bound checks/etc and allow only dereferencing two pointers each loop iteration to do so, basically caching the address of each seq item.
3 participants
No description provided.