init: fixes file descriptor accounting #30065
Conversation
|
The following sections might be updated with supplementary metadata relevant to reviewers and maintainers. Code CoverageFor detailed information about the code coverage, see the test coverage report. ReviewsSee the guideline for information on the review process.
If your review is incorrectly listed, please react with 👎 to this comment and the bot will ignore it on the next update. ConflictsReviewers, this pull request conflicts with the following ones:
If you consider this pull request important, please also help to review the conflicting pull requests. Ideally, start with the one that should be merged first. |
src/init.cpp
Outdated
| nMaxConnections = std::max(std::min<int>(nMaxConnections, fd_max - nBind - MIN_CORE_FILEDESCRIPTORS - MAX_ADDNODE_CONNECTIONS - NUM_FDS_MESSAGE_CAPTURE), 0); | ||
| if (nFD < MIN_CORE_FILEDESCRIPTORS) | ||
| return InitError(_("Not enough file descriptors available.")); | ||
| nMaxConnections = std::min(nFD - MIN_CORE_FILEDESCRIPTORS - MAX_ADDNODE_CONNECTIONS - NUM_FDS_MESSAGE_CAPTURE, nMaxConnections); |
There was a problem hiding this comment.
Notice how the latter nMaxConnections trim is redundant once we limit nFD = std::min(FD_SETSIZE, nFD);. Moving things around, plus accounting for nBind (which is currently not being accounted for in master), we get:
nMaxConnections = std::max(std::min<int>(nMaxConnections, nFD - nBind - MIN_CORE_FILEDESCRIPTORS - MAX_ADDNODE_CONNECTIONS - NUM_FDS_MESSAGE_CAPTURE), 0);
nMaxConnections = std::min(nMaxConnections, nFD - nBind - MIN_CORE_FILEDESCRIPTORS - MAX_ADDNODE_CONNECTIONS - NUM_FDS_MESSAGE_CAPTURE,);
|
Concept ACK |
|
I think it is worth mentioning that the current approach is defining the minimum amount of file descriptions required without accounting for a single connection*, this means that if we are at the bare minimum, we will run but won't be able to connect to any nodes. This is consistent with the current logic in master, but I think it's not the way it should be. I'm happy to add another commit addressing this, but I've rathered start approaching this sticking to the same assumptions as master. * This actually accounts for manual connections, which may never happen, but not to any of our outbound. If we happen to not create any manuals we'd have 8 slots ( |
Concept ACK, I stumbled on this recently in #29415 |
We are computing our file descriptors limits based on whether we use poll or select. However, we are taking that into account only partially (subtracting from fd_max in one case, but from nFD later on). Moreover, nBind is also only accounted for partially. Simplify and fix this logic
|
Thanks for the suggestions @vasild, I've partially covered them and commented on some of the pending ones |
|
🚧 At least one of the CI tasks failed. Make sure to run all tests locally, according to the Possibly this is due to a silent merge conflict (the changes in this pull request being Leave a comment here, if you need help tracking down a confusing failure. |
sr-gi
force-pushed
the
2024-05-fdcount
branch
2 times, most recently
from
fe92353
to
fa42bf7
Compare
There was a problem hiding this comment.
ACK fa42bf7
This is already an improvement over master. Would be nice to resolve #30065 (comment) as well.
sr-gi
force-pushed
the
2024-05-fdcount
branch
3 times, most recently
from
8ed9261
to
f569b06
Compare
Addressed all comments now |
The current logic for file descriptor accounting is pretty convoluted and hard to follow. This is partially caused by the lack of documentation plus non-intuitive variable naming (which was more intuitive when fewer things were accounted for, but hasn't aged well). This has led to this accounting being error-prone and hard to maintain (as shown in he previous commit). Redefine some of the constants, plus document what are we accounting for so this can be extended more easily Remove FreeBSD workaround to bitcoin#2695
|
Nice catch @vasild, I've squashed that into one of the previous commits. It should be good now. |
The current logic for file descriptor accounting is pretty convoluted and hard to follow. This is partially caused by the lack of documentation plus non-intuitive variable naming (which was more intuitive when fewer things were accounted for, but
hasn't aged well). This has led to this accounting being error-prone and hard to maintain (as shown in the first commit of this PR).
Redefine some of the constants, plus document what are we accounting for so this can be extended more easily
Fixes #18911