♻️ Make session manager injectable and isolate for v7 rewrite

[thomas-lebeau]

Motivation

Prepare the session manager layer for a from-scratch rewrite in v7. The existing session manager internals (cookies, storage, core session store) are tightly coupled to boot-level tests, making it hard to replace the implementation without breaking unrelated tests.

This PR decouples all non-session-manager code from session internals, so the session manager can be rewritten in isolation.

Changes

1. Injectable session manager factory

Move the stub-vs-real session manager branching from createPreStartStrategy to the caller (rumPublicApi / logsPublicApi) via a factory parameter:

graph LR
    A[rumPublicApi / logsPublicApi] -->|ternary: bridge ? stub : real| B[factory]
    B -->|passed as param| C[createPreStartStrategy]
    C -->|calls factory| D[session manager instance]

Loading

• BREAKING CHANGE: When no storage strategy is available (cookies/localStorage unavailable), neither SDK will send data. Previously Logs would still start without storage — now it requires explicit sessionPersistence: 'memory' to work without storage. In worker environments, Logs uses the real session manager with memory persistence (we can revisit that when we star using CookieStore as this seems supported in service workers). RUM is explicitly blocked with a warning.
• createPreStartStrategy receives one factory, just calls it — no branching logic
• Callers use: canUseEventBridge() ? stub : real

2. Prove isolation via gut-then-restore

To verify that all non-session-manager tests use mocks (not real implementations), the session manager bodies were temporarily gutted and their unit tests skipped (xdescribe). All 3071 remaining tests passed with zero failures, proving the isolation is complete. The implementations and tests were then fully restored.

3. Mock infrastructure

• Boot tests (preStartRum.spec, preStartLogs.spec) inject mock factories — no more stopSessionManager() cleanup or cookie manipulation
• Public API tests (rumPublicApi.spec, logsPublicApi.spec) inject mock session managers via optional parameter on makeRumPublicApi / makeLogsPublicApi
• mockRumSessionManager / mockLogsSessionManager updated with factory wrappers

4. Worker environment support

Use a defense-in-depth strategy for worker environments, handling workers at multiple layers:

• Storage layer: selectSessionStoreStrategyType forces SessionPersistence.MEMORY in workers — cookies and localStorage are unavailable in worker contexts
• Session manager: Skip DOM event tracking (trackActivity, trackVisibility, trackResume) in workers — no DOM events to listen to
• RUM: Block entirely in preStartRum with an explicit warning — RUM features (DOM observation, user interactions) are not applicable in workers
• BREAKING CHANGE: Logs Uses the real session manager (not a stub) — the storage and session manager layers handle the worker context, so Logs works normally with memory-backed sessions. this means that now logs will have a proper session Id

New exports from @datadog/browser-core: selectSessionStoreStrategyType, SessionStoreStrategyType
startSessionManager now takes sessionStoreStrategyType as a direct parameter.

Checklist

• Tested locally
• Tested on staging
• Added unit tests for this change.
• Added e2e/integration tests for this change.
• Updated documentation and/or relevant AGENTS.md file

[cit-pr-commenter-54b7da]

Bundles Sizes Evolution

📦 Bundle Name	Base Size	Local Size	𝚫	𝚫%	Status
Rum	169.85 KiB	169.95 KiB	+109 B	+0.06%	✅
Rum Profiler	4.29 KiB	4.29 KiB	0 B	0.00%	✅
Rum Recorder	24.71 KiB	24.71 KiB	0 B	0.00%	✅
Logs	56.87 KiB	56.89 KiB	+20 B	+0.03%	✅
Flagging	944 B	944 B	0 B	0.00%	✅
Rum Slim	126.47 KiB	126.56 KiB	+91 B	+0.07%	✅
Worker	23.63 KiB	23.63 KiB	0 B	0.00%	✅

🚀 CPU Performance

Action Name	Base CPU Time (ms)	Local CPU Time (ms)	𝚫%
RUM - add global context	0.0047	0.0045	-4.26%
RUM - add action	0.0133	0.0134	+0.75%
RUM - add error	0.0147	0.0125	-14.97%
RUM - add timing	0.0035	0.0026	-25.71%
RUM - start view	0.0158	0.0121	-23.42%
RUM - start/stop session replay recording	0.001	0.0007	-30.00%
Logs - log message	0.0191	0.0127	-33.51%

🧠 Memory Performance

Action Name	Base Memory Consumption	Local Memory Consumption	𝚫
RUM - add global context	28.42 KiB	27.61 KiB	-832 B
RUM - add action	103.20 KiB	91.88 KiB	-11.32 KiB
RUM - add timing	26.37 KiB	26.92 KiB	+556 B
RUM - add error	103.65 KiB	104.09 KiB	+453 B
RUM - start/stop session replay recording	25.65 KiB	25.97 KiB	+326 B
RUM - start view	474.72 KiB	461.18 KiB	-13.54 KiB
Logs - log message	46.89 KiB	46.57 KiB	-322 B

🔗 RealWorld

[datadog-datadog-prod-us1]

✅ Tests

🎉 All green!

❄️ No new flaky tests detected
🧪 All tests passed

🎯 Code Coverage (details)
• Patch Coverage: 77.78%
• Overall Coverage: 77.32% (+0.00%)

_{This comment will be updated automatically if new data arrives.
🔗 Commit SHA: 8e76c0c | Docs | Datadog PR Page | Was this helpful? Give us feedback!}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: e356ec178a

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[mormubis]

Enabling the session manager to work on service workers seem to increase the complexity.

💬 suggestion: Should we move the trackActivity/trackVisibility/trackResume to an upper level, an call sessionManager.expandOrRenewSession?‏

[thomas-lebeau]

The complexity was already here, it has just been moved from preStartRum and preStartLogs to the session Manager (we were using a stub session Manager instead of the real one)

Moving that back one level up will risk duplicating the logic (LOGS vs RUM, even though RUM does not work in worker env). I think it's fine, so far it's just one if statement but we will need to keep an eye on this.

In v8, is we manage to reconcile session manager and only have one instance, we might want to do differently then.

I will keep note on this though, maybe we can be smarter and guard by feature instead of isWorkerEnvironment (i.e. document is present,...)