Add Scan and Lookup Traces. #2304
Merged
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
batiati
commented
Sep 11, 2024
Comment on lines
-40
to
-56
| /// Since the workload is always sorted by timestamp, | ||
| /// adjacent objects are often going to be in the same table-value block. | ||
| /// The grid is aware when N lookups ask for the same grid block concurrently, | ||
| /// and queues up the reads internally such that they actually hit the storage once. | ||
| /// | ||
| /// To maximize IO utilization, we allow at least `Grid.read_iops_max` lookups to run, | ||
| /// up to an arbitrary constant based on the maximum number of objects per block. | ||
| /// Reasoning: the larger the block size is, higher is the probability of multiple | ||
| /// lookups hitting the same grid block: | ||
| const lookup_workers_max = @max( | ||
| stdx.div_ceil( | ||
| @divFloor(constants.block_size, @sizeOf(Object)), | ||
| Grid.read_iops_max, | ||
| ), | ||
| Grid.read_iops_max, | ||
| ); | ||
|
|
There was a problem hiding this comment.
This logic was removed:
- It simplified the mental model of always handling the max of
Grid.read_iops_maxparallel lookups. - Also allowed using the same
traceevent for lookups from prefetch and scans. - In practice, the benchmark didn't show any performance improvement.
- Reducing the number of
[max]LookupWorkerin the array also reduced the size of the binary 🤯
src/trace.zig
Outdated
Comment on lines
203
to
211
| inline else => |data| { | ||
| comptime assert(@TypeOf(data) == void); | ||
| return event_stack_base.get(event.*); | ||
| }, |
There was a problem hiding this comment.
How about this:
Suggested change
| inline else => |data| { | |
| comptime assert(@TypeOf(data) == void); | |
| return event_stack_base.get(event.*); | |
| }, | |
| inline else => |data, event_tag| { | |
| comptime assert(@TypeOf(data) == void); | |
| return comptime event_stack_base.get(event_tag); | |
| }, |
Iiuc we use inline else rather than else so that we can assert @TypeOf(data) == void.
So since we are generating code for each variant anyway, may as well inline the stack base as a constant rather than needing to fetch it at runtime from the EnumArray.
| return stack_base + @as(u32, @intCast(data.iop)); | ||
| }, | ||
| inline else => |data| { | ||
| comptime assert(@TypeOf(data) == void); |
There was a problem hiding this comment.
Also add comptime assert(event_stack_cardinality.get(event_tag) == 1);.
| levels: [constants.lsm_levels]LevelBuffer, | ||
|
|
||
| pub fn init(self: *ScanBuffer, allocator: Allocator) !void { |
There was a problem hiding this comment.
Maybe pull index into an options struct, so that the call site is labeled?
- Lookup/LookupWorker - ScanTree/ScanTreeLevel
batiati
force-pushed
the
batiati-trace
branch
from
5486af4
to
b2a8b6c
Compare
sentientwaffle
approved these changes
Sep 13, 2024
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.
Add new Traces:
Note 1: Spall sorts the "Thread ID" alphabetically, so it fails to render a usable visualization of this trace, as
ScanTreeandScanTreeLevelare nested events (e.g., eachScanTreehas up tolsm_levelsScanTreeLevel) and stack number (displayed as "Thread ID: ##") is the only thing that groups them together.Note 2: For a more "illustrated" trace sample, try running the benchmark with the flags
--id-order=randomto skip the negative lookup optimization and--transfer-pendingto scan by expired transfers.