bd-workflows: implement type-based routing within the workflow engine by snowp · Pull Request #577 · bitdriftlabs/shared-core

snowp · 2026-08-07T12:58:49Z

Instead of always evaluating all workflows for all logs, this tracks a running set of candidate workflows based
on the log type of the log. This allows us to bypass workflows whose next match transitions require a different
log type than the inbound log.

We keep a router table that provides a routing layer on top of the simple workflow list: for workflows that can be determined to require a particular log type we can filter down the considered workflows to workflows constrained to a log type + all ambiguous workflows. This is kept up to date during workflow execution, rebuilding the routing rules as the list of active runs.

For now this is only going to apply to initial conditions due to not working with the run timeout defined on most workflows, but the initial run ends up being a very significant part of the workflows considered in practice.

Fixes BIT-9477

Copilot

Pull request overview

Adds log-type-aware workflow routing to avoid evaluating workflows that cannot match an incoming log.

Changes:

Adds conservative matcher log-type inference.
Introduces and integrates a workflow candidate router.
Adds routing and integration tests.

Reviewed changes

Copilot reviewed 11 out of 11 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
`bd-workflows/src/workflow.rs`	Computes routes from active workflow state.
`bd-workflows/src/test.rs`	Supports typed test logs.
`bd-workflows/src/routing.rs`	Implements candidate routing.
`bd-workflows/src/routing_test.rs`	Tests route selection and refreshes.
`bd-workflows/src/lib.rs`	Registers the routing module.
`bd-workflows/src/engine.rs`	Integrates routing into event processing.
`bd-workflows/src/engine_test.rs`	Tests routing across workflow transitions.
`bd-workflows/src/engine_test_helpers.rs`	Propagates test log types.
`bd-workflows/src/config.rs`	Aggregates configured log types.
`bd-log-matcher/src/matcher.rs`	Adds log-type sets and matcher inference.
`bd-log-matcher/src/matcher_test.rs`	Tests type sets and inference.

Copilot

Pull request overview

Copilot reviewed 10 out of 10 changed files in this pull request and generated no new comments.

Suppressed comments (1)

bd-workflows/src/routing.rs:186

Vec::reserve takes an additional count relative to len, not a target-capacity delta. Because these vectors are cleared before this helper runs, growing from (for example) capacity 4 to 6 calls reserve(2), which can do nothing and later forces candidate_indices to allocate while selecting a log. Reserve from the current length so prepare actually preserves the documented allocation-free processing path.

    values.reserve(capacity - values.capacity());

mattklein123

How can we understand the perf improvment or not for this in different cases? I think you added a bench can we run it against this change in different scenarios and show the deltas?

mattklein123 · 2026-08-11T17:59:11Z

+          );
+        }
+      },
+      WorkflowEvent::SessionStart(_) | WorkflowEvent::StateChange(..) => {


At least for state change can't we also filter based on whether a workflow cares about it or not? Not sure about SessionStart but I think in that case we could do it also?

Yeah i did this, it was a pretty easy extension

mattklein123 · 2026-08-11T18:00:17Z

+// TODO(snowp): We should be able to apply this routing logic to other event types to dramatically
+// reduce the number of workflows attempted for state events. For now state events are rare so we
+// limit this to logs.


I commented on this above.

mattklein123 · 2026-08-11T18:01:36Z

+/// This is a derived view over the engine's primary workflow vector: every stored index refers to
+/// the workflow at that same position. The caller must rebuild it after adding, removing, or
+/// reordering workflows; between those boundaries, the router updates only workflows which
+/// processed the current log.


nit: not processed, but actual advanced 1 or more traversals?

mattklein123 · 2026-08-11T18:05:01Z

+          .unwrap_or_default()
+      });
+    } else {
+      Self::rebuild_log_routes(log_router, workflows, configs);


Per other comments would be nice to now do this always for the other events, even if rare.

mattklein123 · 2026-08-11T18:07:04Z

+      configs,
+      stats,
+      needs_state_persistence,
+      log_router,


It seems a little fragile to me that we mutate the router in place and use that mutation for rebuilds, but then could theoretically early exit. Can we somehow debug_assert at least that we finalize anything that should be finalized in a single pass?

mattklein123 · 2026-08-11T18:08:39Z

+      log_router.refresh_selected_routes(|index| {
+        workflows
+          .get(index)
+          .zip(configs.get(index))
+          .map(|(workflow, config)| workflow.log_route(config))
+          .unwrap_or_default()
+      });


This is not cheap even in the case where nothing advanced. Can we improve the performance of the very common no advance case?

I tried this out and it seems like for the large corpus that adding an additional value that we track whether we can safely skip refresh ends up being more expensive than just doing it. Tracking the additional value added ~5% over just doing the route rebuild

snowp · 2026-08-12T18:00:53Z

Running it against the internal corpus I got very good results:

origin/main | 114.58 ms
Latest routing changes | 59.82 ms
Change | 47.8% faster

This is primarily from this being a low to zero match session. I'll figure out how to include this larger corpus and we could formalize a way to report on workflow engine changes

snowp · 2026-08-13T03:19:09Z

/benchmark workflows

Instead of always evaluating all workflows for all logs, this tracks a running set of candidate workflows based on the log type of the log. This allows us to bypass workflows whose next match transitions require a different log type than the inbound log

Co-Authored-By: GPT-5 <codex@openai.com>

Route logs by inferred log type and route state changes and session starts to workflows whose active transitions can consume them. Keep the index-based router synchronized with the workflow vector, refresh only selected routes, and cover bucket updates and fallback behavior. Co-Authored-By: GPT-5 <codex@openai.com>

Pin the oversized async log buffer future in the test task to satisfy Clippy large_futures. Co-Authored-By: GPT-5 <codex@openai.com>

Co-Authored-By: GPT-5 <codex@openai.com>

snowp · 2026-08-13T14:04:25Z

/benchmark workflows

github-actions · 2026-08-13T14:09:57Z

Benchmark: `workflows` passed

Compared 8f91cabeba6f20791ccbeb098a42251cb4757258 with merge base 67094ded51afd3959e6c4dd0fc5e77e65b59ad81 (PR base 67094ded51afd3959e6c4dd0fc5e77e65b59ad81).

Criterion mean change

Benchmark	Absolute	Percent
`workflow_replay/large`	+105.86 ms	+38.39%
`workflow_replay/small`	+1.22 µs	+10.02%

The Criterion HTML report is attached to this workflow run.

snowp requested a balanced review from Copilot August 7, 2026 12:58

Copilot started reviewing on behalf of snowp August 7, 2026 12:59 View session

Copilot AI reviewed Aug 7, 2026

View reviewed changes

snowp force-pushed the sp/workflow-log-type-routing branch from c0456d0 to 50f18f3 Compare August 11, 2026 15:13

snowp requested a balanced review from Copilot August 11, 2026 16:10

Copilot started reviewing on behalf of snowp August 11, 2026 16:11 View session

Copilot AI reviewed Aug 11, 2026

View reviewed changes

mattklein123 reviewed Aug 11, 2026

View reviewed changes

snowp force-pushed the sp/workflow-log-type-routing branch from 2eb4522 to 9bb1692 Compare August 13, 2026 03:18

snowp and others added 19 commits August 13, 2026 10:03

Fix workflow routing test lint

02b42ff

Co-Authored-By: GPT-5 <codex@openai.com>

Track processed debug workflows

dc80326

Co-Authored-By: GPT-5 <codex@openai.com>

Document workflow duration routing limitation

c7b1aae

Co-Authored-By: GPT-5 <codex@openai.com>

comment changes

f6b354b

license headers

61ecf8f

comments

3b3834b

Use fixed log type routing buckets

cd1acbb

Co-Authored-By: GPT-5 <codex@openai.com>

Test log type bucket indices

55fdd20

Co-Authored-By: GPT-5 <codex@openai.com>

Document workflow router index invariants

61f70c3

Co-Authored-By: GPT-5 <codex@openai.com>

Document log routing allocation tradeoff

a5b30c7

Co-Authored-By: GPT-5 <codex@openai.com>

Clarify workflow router route insertion

67961ae

Co-Authored-By: GPT-5 <codex@openai.com>

comment

3ab75fb

optimize replacement, remove manual sort

4a757f5

fmt

cfff392

Avoid unchecked workflow route bucket access

1babc59

Co-Authored-By: GPT-5 <codex@openai.com>

clippy

5398a0e

Fix async log buffer test future lint

ebafe3b

Pin the oversized async log buffer future in the test task to satisfy Clippy large_futures. Co-Authored-By: GPT-5 <codex@openai.com>

snowp and others added 4 commits August 13, 2026 10:03

Assert workflow event routes are finalized once

5ce2a25

Co-Authored-By: GPT-5 <codex@openai.com>

Assert workflow route builders finish

fb4f61e

Co-Authored-By: GPT-5 <codex@openai.com>

Simplify workflow route builder finalization

1cbb6ee

Co-Authored-By: GPT-5 <codex@openai.com>

Clarify workflow route finalization assertion

8f91cab

Co-Authored-By: GPT-5 <codex@openai.com>

snowp force-pushed the sp/workflow-log-type-routing branch from 9bb1692 to 8f91cab Compare August 13, 2026 14:04

Conversation

snowp commented Aug 7, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

mattklein123 left a comment

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!

snowp commented Aug 12, 2026

Uh oh!

snowp commented Aug 13, 2026

Uh oh!

snowp commented Aug 13, 2026

Uh oh!

github-actions Bot commented Aug 13, 2026

Benchmark: workflows passed

Criterion mean change

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

snowp commented Aug 7, 2026 •

edited

Loading

Benchmark: `workflows` passed