Skip to content

Conversation

@lukevalenta
Copy link
Contributor

@lukevalenta lukevalenta commented May 22, 2025

c913c38 Fix comment about dropped channels

The value in the channel will persist even after the sender is dropped,
so remove the comment. According to
https://docs.rs/tokio/latest/tokio/sync/watch/index.html#closing, "The
value in the channel will not be dropped until the sender and all
receivers have been dropped."

f721b0c Add breakdown of full vs partial tiles to metrics

a053648 Add ability to set a maximum timeout for skipped entries

When a pending entry has been queued for longer than the timeout, it
will be included in the next sequencing regardless of how many times it
has exceeded the number of allowed skipped sequencings.

859df92 Add option to disable deduplication checks

Add an option to the configuration to disable the deduplication cache,
which is useful for running tests and benchmarks against a log.

a16ee94 Add batcher parameters to log configuration, and allow setting zero batchers

  • Add ability to set num_batchers to zero so the frontend worker
    connects directly to the sequencer.
  • Remove MAX_IN_FLIGHT check from batcher. If we need some sort of
    rate-limiting in the batchers, we can add it later based on
    operational experience.

0f92d2a Allow skipping checkpoint creation when there are no entries to sequence

  • Restore old log behavior by setting default 'max_sequence_skips' to 0
  • Rework PoolState::take() to return an Option indicating whether or not
    we sequencing can be skipped are no new entries or entries that don't
    make up a full tile.
  • Set additionalProperties:false in JSON schema

e9d76c0 Specify sequence interval in milliseconds instead of seconds

@lukevalenta lukevalenta requested a review from cjpatton May 22, 2025 19:13
@lukevalenta lukevalenta self-assigned this May 22, 2025
- Restore old log behavior by setting default 'max_sequence_skips' to 0
- Rework PoolState::take() to return an Option indicating whether or not
  we sequencing can be skipped are no new entries or entries that don't
  make up a full tile.
- Set additionalProperties:false in JSON schema
…atchers

- Add ability to set num_batchers to zero so the frontend worker
  connects directly to the sequencer.
- Remove MAX_IN_FLIGHT check from batcher. If we need some sort of
  rate-limiting in the batchers, we can add it later based on
  operational experience.
Add an option to the configuration to disable the deduplication cache,
which is useful for running tests and benchmarks against a log.
@lukevalenta lukevalenta force-pushed the lvalenta/tuning branch 2 times, most recently from 8a44f7c to ded12d5 Compare June 4, 2025 13:58
When a pending entry has been queued for longer than the timeout, it
will be included in the next sequencing regardless of how many times it
has exceeded the number of allowed skipped sequencings.
The value in the channel will persist even after the sender is dropped,
so remove the comment.  According to
https://docs.rs/tokio/latest/tokio/sync/watch/index.html#closing, "The
value in the channel will not be dropped until the sender and all
receivers have been dropped."
Copy link
Contributor

@cjpatton cjpatton left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Still working my way through the stack, but I got to a point where I have question I want to answer before I keep looking.

Copy link
Contributor

@rozbb rozbb left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

nice!

@lukevalenta lukevalenta merged commit 654ca6a into main Jun 5, 2025
1 check passed
@lukevalenta lukevalenta deleted the lvalenta/tuning branch June 5, 2025 14:01
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants