CDFs for EB announcements by coot · Pull Request #6639 · IntersectMBO/cardano-node

coot · 2026-08-05T09:51:53Z

Description

CDF counters - refactorisation
CDFs for EB announcements

Checklist

Commit sequence broadly makes sense and commits have useful messages
New tests are added if needed and existing tests are updated. These may include:
- golden tests
- property tests
- roundtrip tests
- integration tests
  See Running tests for more details
Any changes are noted in the CHANGELOG.md for affected package
- cardano-node-chairman, cardano-submit-api and cardano-testnet instead need a
  changelog fragment in <package>/.changes/, because their CHANGELOG.md is generated
  from fragments at release time. Copy _TEMPLATE.yml from that directory, or run
  nix run github:input-output-hk/cardano-dev#herald -- new
The version bounds in .cabal files are updated
CI passes. See note on CI. The following CI checks are required:
- Code is linted with hlint. See .github/workflows/check-hlint.yml to get the hlint version
- Code is formatted with stylish-haskell. See .github/workflows/stylish-haskell.yml to get the stylish-haskell version
- Code builds on Linux, MacOS and Windows for ghc-9.6 and ghc-9.12
Self-reviewed the diff

Note on CI

If your PR is from a fork, the necessary CI jobs won't trigger automatically for security reasons.
You will need to get someone with write privileges. Please contact IOG node developers to do this
for you.

* Moved to a new module * Provide re-usable API This is a preparation work for re-using the CDF calculations for Leios.

nfrisby · 2026-08-05T15:09:07Z

+
+data Config = Config { numOfDataPoints :: Int }
+
+-- | Default `Config` keeps `k/2` data points.


The proportionality to k is easier to justify for some CDFs than for others, but it's probably fine for most?

But one other option is to just force each callers to choose a number

nfrisby · 2026-08-05T15:12:20Z

+  -- ^ index, priority, value
+  -> State p
+  -> f Counter
+  -> (f Counter, State p, Bool)


A couple questions that a comment should answer:

What does the Bool mean?

Why do we need to maintain the Counter and the State as separate things?

There's only two call sites to this function and they both ignore the return value when the returned Bool is False.

So return Maybe (f Counter, State p) instead?

nfrisby · 2026-08-05T15:12:53Z

+    = if Pq.size m' > numOfDataPoints
+      then
+        case Pq.minView m' of
+          Nothing -> (cdfs, s, False)


use nothingToDo here too?

nfrisby · 2026-08-05T15:13:31Z

+  where
+    nothingToDo = (cdfs, s, False)
+
+    delay_ :: Double


Just inline this as (realToFrac delay :: Double)?

nfrisby · 2026-08-05T15:15:26Z

+  | otherwise = cdf
+
+
+newtype State p = State { cdfState :: IntPSQ p NominalDiffTime }


Please add a comment explaining why a Cdf needs a notion of priority?

nfrisby · 2026-08-05T15:42:08Z

+data LeiosMetrics' c = LeiosMetrics {
+    lmCdfState     :: Cdf.State SlotNo
+    -- ^ internal state which collects all data points
+  , lmSlotMapLate  :: IntPSQ SlotNo ()


Why isn't this just another c?

And maybe (Maybe SlotNo, Int) would be a simpler data structure to understand?

nfrisby · 2026-08-05T15:43:41Z

+    where
+      size                   = Cdf.size lmCdfState
+      cdfMetric name var     = DoubleM name (fromIntegral (Cdf.counter var) / fromIntegral size)
+      lateAnnoucementsAllMetric = [ CounterM "leios.eb.announcement.late.all" Nothing | lmDelay > 2 ]


Typos:

lateAnnoucementsAllMetric -> lateAnnoumcementsAllMetric

lateAnnoucementsPerSlotMetric -> lateAnnoucementsPerSlotMetric

nfrisby · 2026-08-05T15:50:40Z

+
+calculateLeiosMetrics lm _lc _ = lm
+
+instance LogFormatting LeiosMetrics where


👍 matches

cardano-node/cardano-node/src/Cardano/Node/Tracing/Tracers/Consensus.hs

Lines 535 to 552 in ee7ea74

instance LogFormatting ClientMetrics where

forMachine _dtal _ = mempty

asMetrics ClientMetrics {cmTraceIt = False} = []

asMetrics ClientMetrics {..} =

[ DoubleM "blockfetchclient.blockdelay" cmDelay

, IntM "blockfetchclient.blocksize" (fromIntegral cmBlockSize)

]

++ lateBlockMetric

++ if cmTraceVars

then [ cdfMetric "blockfetchclient.blockdelay.cdfOne" cmCdf1sVar

, cdfMetric "blockfetchclient.blockdelay.cdfThree" cmCdf3sVar

, cdfMetric "blockfetchclient.blockdelay.cdfFive" cmCdf5sVar

]

else []

where

size = Cdf.size cmCdfState

cdfMetric name var = DoubleM name (fromIntegral (Cdf.counter var) / fromIntegral size)

lateBlockMetric = [ CounterM "blockfetchclient.lateblocks" Nothing | cmDelay > 5 ]

nfrisby · 2026-08-05T15:57:54Z

+      , ("leios.eb.announcement.delay.cdf1600ms", "probability for EB announcemement to complete within 1600ms")
+      , ("leios.eb.announcement.delay.cdf1800ms", "probability for EB announcemement to complete within 1800ms")
+      , ("leios.eb.announcement.delay.cdf2000ms", "probability for EB announcemement to complete within 2000ms")
+      , ("leios.e.b.announcement.late.all", "number of late EB announcements that took longer than 2s: all peers are included")


I wasn't expecting "all peers are included" here. These metrics are driven by TraceLeiosAnnouncementAccepted, which is deduped---it only fires the first time some announcement arrives.

nfrisby · 2026-08-05T15:58:35Z

+      , ("leios.eb.announcement.delay.cdf1800ms", "probability for EB announcemement to complete within 1800ms")
+      , ("leios.eb.announcement.delay.cdf2000ms", "probability for EB announcemement to complete within 2000ms")
+      , ("leios.e.b.announcement.late.all", "number of late EB announcements that took longer than 2s: all peers are included")
+      , ("leios.e.b.announcement.late.bySlot", "number of slots for which the earliest EB announcements took longer than 2s")


It looks to me like lmSlotMapLate is counting announcements rather than counting slots. Am I confused?

Overall, I'm not clear on what lmSlotMapLate is actually doing... partly because I'm not sure what IntM is, but if I recall correctly (I can't find its definition right now), it's a gauge, so it's not accumulating anything.

CDF counters - refactorisation

4b6a783

* Moved to a new module * Provide re-usable API This is a preparation work for re-using the CDF calculations for Leios.

coot requested a review from a team as a code owner August 5, 2026 09:51

coot linked an issue Aug 5, 2026 that may be closed by this pull request

Measure diffusion distributions input-output-hk/ouroboros-leios#1026

Open

coot added the leios Means that this tickets is related to the implementation of ouroboros leios. label Aug 5, 2026

CDFs for EB announcements

ee7ea74

coot force-pushed the coot/leios-cdfs branch from 1f7b0f3 to ee7ea74 Compare August 5, 2026 12:59

nfrisby requested changes Aug 5, 2026

View reviewed changes


		data Config = Config { numOfDataPoints :: Int }

		-- \| Default `Config` keeps `k/2` data points.

		\| otherwise = cdf


		newtype State p = State { cdfState :: IntPSQ p NominalDiffTime }


		calculateLeiosMetrics lm _lc _ = lm

		instance LogFormatting LeiosMetrics where

	instance LogFormatting ClientMetrics where
	forMachine _dtal _ = mempty
	asMetrics ClientMetrics {cmTraceIt = False} = []
	asMetrics ClientMetrics {..} =
	[ DoubleM "blockfetchclient.blockdelay" cmDelay
	, IntM "blockfetchclient.blocksize" (fromIntegral cmBlockSize)
	]
	++ lateBlockMetric
	++ if cmTraceVars
	then [ cdfMetric "blockfetchclient.blockdelay.cdfOne" cmCdf1sVar
	, cdfMetric "blockfetchclient.blockdelay.cdfThree" cmCdf3sVar
	, cdfMetric "blockfetchclient.blockdelay.cdfFive" cmCdf5sVar
	]
	else []
	where
	size = Cdf.size cmCdfState
	cdfMetric name var = DoubleM name (fromIntegral (Cdf.counter var) / fromIntegral size)
	lateBlockMetric = [ CounterM "blockfetchclient.lateblocks" Nothing \| cmDelay > 5 ]

Uh oh!

Conversation

coot commented Aug 5, 2026

Description

Checklist

Note on CI

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

nfrisby Aug 5, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

nfrisby Aug 5, 2026 •

edited

Loading