facebook/Haxl
1
1
Fork 0
mirror of https://github.com/facebook/Haxl.git synced 2024-12-24 17:23:03 +03:00
Code Issues Projects Releases Wiki Activity
fdfb86379b
Haxl/changelog.md
Dylan Yudaken fdfb86379b profile more accurately (#119) 
Summary:
Pull Request resolved: https://github.com/facebook/Haxl/pull/119

Currently profiling with labels does not track full stacks.
This is problematic in the case where all top level methods end up calling a common method with a label, as we have no way of attributing to the top level methods.
Eg if A/B/C all call X when running work then we cannot tell which work is expensive, as we have no connection from the work through X

Reviewed By: simonmar

Differential Revision: D20384255

fbshipit-source-id: f9aa0462904c17dee32d37a659b491e8d252d6db
2020-04-24 08:02:53 -07:00

3.4 KiB
Raw Blame History

Changes in version

  • Added fetchBatchId to FetchStats
  • Profiling now tracks full stacks

Changes in version 2.3.0.0

  • Removed FutureFetch

Changes in version 2.2.0.0

  • Use BasicHashTable for the Haxl DataCache instead of HashMap
  • API Changes in: Haxl.Core.DataCache, Haxl.Core.Fetch
  • Removed support for GHC < 8.2

Changes in version 2.1.2.0

  • Add a callgraph reference to 'Env' to record the function callgraph during a computation. The callgraph is stored as an edge list in the Env through the use of withCallGraph and enables users to debug a Haxl computation.

Changes in version 2.1.1.0

  • Adds feature to track outgone datasource fetches. This is only turned on for report level greater than 1. The fetches are stored as a running Map in the env ('submittedReqsRef').

Changes in version 2.1.0.0

  • Add a new 'w' parameter to 'GenHaxl' to allow arbitrary writes during a computation. These writes are stored as a running log in the Env, and are memoized. This allows users to extract information from a Haxl computation which throws. Our advise is to limit these writes to monitoring and debugging logs.

  • A 'WriteTree' constructor to maintain log of writes inside the Environment. This is defined to allow O(1) mappend.

Changes in version 2.0.1.1

  • Support for GHC 8.6.1
  • Bugfixes

Changes in version 2.0.1.0

  • Exported MemoVar from Haxl.Core.Memo
  • Updated the facebook example
  • Fixed some links in the documentation
  • Bump some version bounds

Changes in version 2.0.0.0

  • Completely rewritten internals to support arbitrarily overlapping I/O and computation. Haxl no longer runs batches of I/O in "rounds", waiting for all the I/O to complete before resuming the computation. In Haxl 2, we can spawn I/O that returns results in the background and computation fragments are resumed when the values they depend on are available. See tests/FullyAsyncTest.hs for an example.

  • A new PerformFetch constructor supports the new concurrency features: BackgroundFetch. The data source is expected to call putResult in the background on each BlockedFetch when its result is ready.

  • There is a generic DataSource implementation in Haxl.DataSource.ConcurrentIO for performing each I/O operation in a separate thread.

  • Lots of cleanup and refactoring of the APIs.

  • License changed from BSD+PATENTS to plain BSD3.

Changes in version 0.5.1.0

  • 'pAnd' and 'pOr' were added
  • 'asyncFetchAcquireRelease' was added
  • 'cacheResultWithShow' was exposed
  • GHC 8.2.1 compatibility

Changes in version 0.5.0.0

  • Rename 'Show1' to 'ShowP' (#62)

Changes in version 0.3.0.0

  • Some performance improvements, including avoiding quadratic slowdown with left-associated binds.

  • Documentation cleanup; Haxl.Core is the single entry point for the core and engine docs.

  • (>>) is now defined to be (*>), and therefore no longer forces sequencing. This can have surprising consequences if you are using Haxl with side-effecting data sources, so watch out!

  • New function withEnv, for running a sub-computation in a local Env

  • Add a higher-level memoization API, see 'memo'

  • Show is no longer required for keys in cachedComputation

  • Exceptions now have Eq instances