Fix --block-io #310
Merged
Fix --block-io #310
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
bmario
requested changes
Mar 13, 2024
include/lo2s/monitor/bio_monitor.hpp
Outdated
|
|
||
| BlockDevice dev = block_device_for<RecordBlock>(event); | ||
|
|
||
| if (sector_cache_.count(dev) == 0 && sector_cache_[dev].count(event->sector) == 0) |
There was a problem hiding this comment.
Suggested change
| if (sector_cache_.count(dev) == 0 && sector_cache_[dev].count(event->sector) == 0) | |
| if (sector_cache_.count(dev) == 0 || sector_cache_.at(dev).count(event->sector) == 0 || sector_cache_.at(dev).at(event->sector) == 0 ) |
it can't be both, right?
include/lo2s/perf/bio/writer.hpp
Outdated
|
|
||
| BlockDevice dev = block_device_for<RecordBlock>(event); | ||
|
|
||
| if (sector_cache_.count(dev) == 0 && sector_cache_[dev].count(event->sector) == 0) |
There was a problem hiding this comment.
Suggested change
| if (sector_cache_.count(dev) == 0 && sector_cache_[dev].count(event->sector) == 0) | |
| if (sector_cache_.count(dev) == 0 || sector_cache_.at(dev).count(event->sector) == 0) |
this only not crashes because you use operator[]
include/lo2s/perf/bio/writer.hpp
Outdated
Comment on lines
110
to
114
| if (sector_cache_.count(dev) == 0) | ||
| { | ||
| sector_cache_.emplace(std::piecewise_construct, std::forward_as_tuple(dev), | ||
| std::forward_as_tuple(std::map<uint64_t, uint64_t>())); | ||
| } |
There was a problem hiding this comment.
I don't think this is necessary, as you use operator[] and only default arguments.
include/lo2s/perf/bio/writer.hpp
Outdated
| sector_cache_.emplace(std::piecewise_construct, std::forward_as_tuple(dev), | ||
| std::forward_as_tuple(std::map<uint64_t, uint64_t>())); | ||
| } | ||
| sector_cache_[dev][event->sector] = size; |
There was a problem hiding this comment.
Suggested change
| sector_cache_[dev][event->sector] = size; | |
| sector_cache_[dev][event->sector] += size; |
why not accumulate instead of overwriting?
There was a problem hiding this comment.
Why do we want to accumulate here? We want to cache the size of that specific queue-ing operation here.
There was a problem hiding this comment.
what happens if there's another request for the same sector?
include/lo2s/perf/bio/writer.hpp
Outdated
| handle, size, event->sector); | ||
| handle, sector_cache_[dev][event->sector], | ||
| event->sector); | ||
| sector_cache_[dev].erase(event->sector); |
There was a problem hiding this comment.
Suggested change
| sector_cache_[dev].erase(event->sector); | |
| sector_cache_[dev][event->sector] = 0; |
cvonelm
force-pushed
the
issue-290-block-io-broken
branch
from
39cbc61 to
5c9696f
Compare
bmario
requested changes
Mar 14, 2024
include/lo2s/perf/bio/writer.hpp
Outdated
| uint64_t sector; // the accessed sector on the device | ||
|
|
||
| uint32_t nr_sector; // the number of sector_cache_ written | ||
| // 512) for complete: the error code of the operation |
include/lo2s/perf/bio/writer.hpp
Outdated
| sector_cache_.emplace(std::piecewise_construct, std::forward_as_tuple(dev), | ||
| std::forward_as_tuple(std::map<uint64_t, uint64_t>())); | ||
| } | ||
| sector_cache_[dev][event->sector] = size; |
There was a problem hiding this comment.
what happens if there's another request for the same sector?
bmario
approved these changes
Apr 23, 2024
Due to a bug, when the multi-reader switches from one reader with later events to one that has earlier events, a block I/O event was discarded from the reader with the later events. This predictably led to event loss. Further, the insert event was switched from block_rq_insert to block_bio_queue. The reason for this is that block_rq_insert never matched the amount of data reported by block_rq_issue and block_rq_complete. I could not find another block_rq_* tracepoint that accounts for the block I/O request that do not enter through block_rq_insert. However, block_bio_queue matches the amount of data that block_rq_issue and block_rq_complete report. This results in somewhat of a mismatch, as block_bio_* is one level higher up than block_rq_*. This is most notable in that one block_bio_queue "struct bio" is split into multiple (usually less than 10) "struct rq" on the block_rq_* level. I remedy this situation by only writing the first block_rq_insert/issue event that I encounter (which is the one with the matching sector to the block_bio_queue event) and discarding the others. We might want to discuss if using the timestamp of the last block_rq_* event is the more correct variant.
cvonelm
force-pushed
the
issue-290-block-io-broken
branch
from
931a8f8 to
c7dcbcb
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Due to a bug, when the multi-reader switches from one reader with later events to one that has earlier events, a block I/O event was discarded from the reader with the later events. This predictably led to event loss.
Further, the insert event was switched from block_rq_insert to block_bio_queue. The reason for this is that block_rq_insert never matched the amount of data reported by block_rq_issue and block_rq_complete. I could not find another block_rq_* tracepoint that accounts for the block I/O request that do not enter through block_rq_insert. However, block_bio_queue matches the amount of data that block_rq_issue and block_rq_complete report.
This results in somewhat of a mismatch, as block_bio_* is one level higher up than block_rq_. This is most notable in that one block_bio_queue "struct bio" is split into multiple (usually less than 10) "struct rq" on the block_rq_ level.
I remedy this situation by only writing the first block_rq_insert/issue event that I encounter (which is the one with the matching sector to the block_bio_queue event) and discarding the others.
We might want to discuss if using the timestamp of the last block_rq_* event is the more correct variant.
This Fixes #290