feat: base provider state on CDN fetches only, not SSE connection state by toddbaert · Pull Request #44 · Dynatrace/openfeature-provider-java

toddbaert · 2026-10-19T19:54:49Z

Simplifies SSE, as described here, and update e2e accordingly.

TLDR: if SSE is connected, we poll slow - if SSE is not connected, we poll fast, regardless of the SSE connection being disabled by config or disconnect by accident; only failed polls impact provider state

Spec: https://github.com/Dynatrace-Internal/feature-management-knowledge-base/pull/177
Harness: feat: use fast-poll mode if fetch fails while SSE connected fm-provider-test-harness#21 (released in v1.4.0)
Node: feat: base provider state on CDN fetches only, not SSE connection state openfeature-provider-js#46
Python: feat: introduce SSE (ref AI-984) openfeature-provider-python#28 (draft)

alexandraoberaigner

Still trying to understand the exact flow of the sse connection within the providers, I found 2 spots where we are somewhat 'loosing' conditions with the code removal. I'm not confident that they are really issues, but might be worth looking at again :)

alexandraoberaigner · 2026-10-23T09:22:18Z

+        Thread.sleep(100); // let the interval switch settle
        fetches.set(0);
        Thread.sleep(300);


[q/nit] should those timings be constants to make it clear what they wait for? 100 & 300 seem to be reused intentionally

alexandraoberaigner · 2026-10-23T09:45:00Z


    // header and value are null when the step has no "with ... header ..." suffix
-    @Given("^an initialized, READY provider serving the \"([^\"]*)\" flag configuration(?: with \"([^\"]*)\" header \"([^\"]*)\")?$")
+    @Given("^an initialized, READY provider serving the \"([^\"]*)\" flag configuration(?: with \"([^\"]*)\" header (\"[^\"]*\"|'[^']*'))?$")


[q/nity-nit] this is independent from the sse state stuff, right?

yes, this is an unrelated change in the testbed, enforcing that etag values are quoted (that's just part of HTTP semantics)

alexandraoberaigner · 2026-10-23T13:23:09Z

-        if (sseConnected.get()) {
-            log.warn("poll failed (SSE healthy, staying READY): {}", reason);
-            return;
-        }


[q] removing this part, will result in an immediate stale state if onFetchFailure is called, right? So, we are much more sensitive to issues of the CDN than we were before, when additionally checking for a healthy sse. We do have the http retry, however, calculating 120ms for MAX_TOTAL_BACKOFF_MS and MAX_ATTEMPTS being 2, the cdn cannot have a very long hiccup.
In your experience is this something we should give more time with by increasing the max retries or adding some grace time?

Good catch!

I think we can fix this with one small addition to my simplification: after a failed fetch, poll at 10s until a fetch succeeds... so pollingManager.setInterval(pollIntervalDisconnected) on failure, and restoring pollIntervalConnected on success if SSE is connected.

So whether SSE is connected or not, a failed fetch drops us to the fast cadence. If it's a blip, we recover within ~10s and don't go to a spurious ERROR over. I'm OK with quick STALE transitions; those are just being truthful, and even our internal provider does this due to gRPC stream timeouts.

I'd rather not add more retries per fetch (a fetch has to stay under the 10s poll interval) or more grace time (it would only delay a correct signal).

Opened:

feat: use fast-poll mode if fetch fails while SSE connected fm-provider-test-harness#21

https://github.com/Dynatrace-Internal/feature-management-knowledge-base/pull/177

Overall, the simplification is still much better, it avoids the entire debounce construct and multiple paths for STALE/ERROR - we just leverage an existing construct (poll aggressiveness) to fix your finding.

The new test in the test harness PR failed until I made the fix I proposed above.

Again, nice find! Let me know if you think this covers it.

alexandraoberaigner · 2026-10-23T14:00:53Z

+        log.info("SSE connected; poll interval switched to {}", pollIntervalConnected);
+        pm.setInterval(pollIntervalConnected);
+        // catch up on anything announced while the stream was down; never dropped
+        pm.requestOutOfBandFetch(null);


[q] similar to the other comment: we loose the (!wasConnected) condition for this catch up fetch. What if there is a short sse hiccup, it reconnects immediately, then it will send this fetch unnecessarily, right, because we dont check anymore?

main also fires the catch-up after a short hiccup (hadPendingDisconnect is true when the stream reopens during the "debounce")...

But I think your other point here is valid: #44 (comment)

Copilot

🟡 Changes recommended

SSE reconnection can override fast recovery polling during CDN throttling, delaying recovery until the ten-minute polling tick.

2 open issues

What changed in this PR

This PR makes provider staleness depend on CDN fetch outcomes rather than SSE connection state, while retaining SSE-based polling intervals and reconnect catch-up fetches.

Changes:

Removes SSE disconnect debouncing and disconnect-driven state transitions.
Accelerates polling after CDN failures and restores the connected interval after successful fetches.
Updates unit tests and end-to-end steps for the revised behavior.

File	Description
`src/test/java/com/dynatrace/openfeature/ProviderTestSupport.java`	Removes the obsolete disconnect-debounce constant.
`src/test/java/com/dynatrace/openfeature/e2e/E2eStepDefinitions.java`	Updates SSE steps, header parsing, and event-absence assertions.
`src/test/java/com/dynatrace/openfeature/DynatraceProviderTest.java`	Revises staleness, reconnect, and polling-cadence tests.
`src/main/java/com/dynatrace/openfeature/DynatraceProvider.java`	Bases staleness and recovery polling on CDN fetch outcomes.

_{Review effort: Balanced}

Copilot

🔵 Needs a closer look

An initialization race can overwrite fast recovery polling with the connected interval after a failed SSE catch-up fetch.

1 open issue

Add distinct-interval tests for fetch failure and recovery cadence

1 previously missed issue

In code that hasn't changed since last review

Prevent stale SSE recovery from using the connected polling interval

src/main/java/com/dynatrace/openfeature/DynatraceProvider.java:557

If an SSE catch-up fetch fails after pollingManager.start() but before initialization adopts the connected cadence (lines 337–340), that handoff can overwrite this fast interval even though the provider is now STALE. Recovery then waits for the 10-minute connected interval, beyond the grace period. Synchronize the initialization handoff with these callbacks and select the connected interval only while READY.

1 resolved since last review

Keep fast polling until provider recovers from CDN failure

_{Review effort: Balanced}

Copilot

🟢 Approval recommended

The reviewed changes address the prior recovery-cadence concern, and no unresolved blocking issues were identified.

No open issues 🎉

1 resolved since last review

Add distinct-interval tests for fetch failure and recovery cadence

_{Review effort: Balanced}

Copilot

🟢 Approval recommended

Static review found no unresolved blocking issues in the state transitions, polling behavior, or test adaptations; tests were not executed.

No open issues 🎉

_{Review effort: Balanced}

Copilot

🟢 Approval recommended

No active review issues are reported for the update to CDN-fetch-driven staleness.

No open issues 🎉

_{Review effort: Balanced}

Signed-off-by: Todd Baert <todd.baert@dynatrace.com>

Copilot

🟢 Approval recommended

There are no active review issues for the changes to CDN-driven staleness and SSE-controlled polling.

No open issues 🎉

_{Review effort: Balanced}

toddbaert mentioned this pull request Oct 19, 2026

feat: SSE state no longer drives provider staleness Dynatrace/fm-provider-test-harness#17

Merged

toddbaert requested review from alexandraoberaigner, jonathannorris, kaushalkapasi and leakonvalinka and removed request for jonathannorris October 19, 2026 20:34

toddbaert commented Oct 19, 2026

View reviewed changes

Comment thread .gitmodules Resolved

toddbaert marked this pull request as ready for review October 20, 2026 17:51

alexandraoberaigner reviewed Oct 23, 2026

View reviewed changes

toddbaert mentioned this pull request Oct 23, 2026

feat: use fast-poll mode if fetch fails while SSE connected Dynatrace/fm-provider-test-harness#21

Merged

Copilot AI balanced review requested due to automatic review settings October 27, 2026 13:44

Copilot started reviewing on behalf of toddbaert October 27, 2026 13:45

View session

Copilot AI reviewed Oct 27, 2026

View reviewed changes

Comment thread src/main/java/com/dynatrace/openfeature/DynatraceProvider.java Resolved

Comment thread src/main/java/com/dynatrace/openfeature/DynatraceProvider.java Resolved

Copilot AI balanced review requested due to automatic review settings October 27, 2026 14:46

Copilot started reviewing on behalf of toddbaert October 27, 2026 14:47

View session

Copilot AI reviewed Oct 27, 2026

View reviewed changes

Copilot AI balanced review requested due to automatic review settings October 27, 2026 15:18

Copilot started reviewing on behalf of toddbaert October 27, 2026 15:18

View session

Copilot AI reviewed Oct 27, 2026

View reviewed changes

Copilot AI balanced review requested due to automatic review settings October 27, 2026 18:06

Copilot started reviewing on behalf of toddbaert October 27, 2026 18:07

View session

Copilot AI reviewed Oct 27, 2026

View reviewed changes

This was referenced Oct 27, 2026

feat: base provider state on CDN fetches only, not SSE connection state Dynatrace/openfeature-provider-js#46

Merged

feat: introduce SSE (ref AI-984) Dynatrace/openfeature-provider-python#28

Open

toddbaert requested review from alexandraoberaigner and a balanced review from Copilot October 27, 2026 20:29

Copilot started reviewing on behalf of toddbaert October 28, 2026 16:28

View session

Copilot AI reviewed Oct 28, 2026

View reviewed changes

feat: base staleness on CDN fetches only, not SSE state

d860d5b

Signed-off-by: Todd Baert <todd.baert@dynatrace.com>

toddbaert added 11 commits November 2, 2026 14:59

fixup: absorb the SSE staleness harness scenarios

3d7e8f9

Signed-off-by: Todd Baert <todd.baert@dynatrace.com>

fixup: assert a disconnect emits no PROVIDER_STALE

6150595

Signed-off-by: Todd Baert <todd.baert@dynatrace.com>

fixup: bump harness

41cd467

Signed-off-by: Todd Baert <todd.baert@dynatrace.com>

fixup: bump harness, bind the GRACE_PERIOD step

625ad32

Signed-off-by: Todd Baert <todd.baert@dynatrace.com>

fixup: pin harness to v1.3.0

84a81ce

Signed-off-by: Todd Baert <todd.baert@dynatrace.com>

fixup: bump harness to v1.3.1, accept single-quoted header values

0be6436

Signed-off-by: Todd Baert <todd.baert@dynatrace.com>

fixup: use fast-poll mode if fetch fails while SSE connected

ef24b46

Signed-off-by: Todd Baert <todd.baert@dynatrace.com>

fixup: only relax poll cadence on SSE connect if READY

3b5eb08

Signed-off-by: Todd Baert <todd.baert@dynatrace.com>

fixup: relax poll cadence only on a successful fetch

398a6cb

Signed-off-by: Todd Baert <todd.baert@dynatrace.com>

fixup: bump harness to v1.4.0

3a1dead

Signed-off-by: Todd Baert <todd.baert@dynatrace.com>

fixup: start SSE after polling mgr

cd94c61

Signed-off-by: Todd Baert <todd.baert@dynatrace.com>

toddbaert force-pushed the draft/sse-independent-staleness branch from b8c3835 to cd94c61 Compare November 2, 2026 19:59

Copilot AI balanced review requested due to automatic review settings November 2, 2026 19:59

Copilot started reviewing on behalf of toddbaert November 2, 2026 20:00

View session

Copilot AI reviewed Nov 2, 2026

View reviewed changes

toddbaert changed the title ~~feat: base staleness on CDN fetches only, not SSE state~~ feat: base provider state on CDN fetches only, not SSE connection state Nov 3, 2026

JamieSinn approved these changes Nov 4, 2026

View reviewed changes

toddbaert merged commit dd3b485 into main Nov 4, 2026
16 checks passed

toddbaert deleted the draft/sse-independent-staleness branch November 4, 2026 21:39

github-actions Bot mentioned this pull request Nov 4, 2026

chore(main): release 0.6.0-alpha #33

Open

Conversation

toddbaert commented Oct 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

alexandraoberaigner left a comment

Choose a reason for hiding this comment

Uh oh!

alexandraoberaigner Oct 23, 2026

Choose a reason for hiding this comment

Uh oh!

alexandraoberaigner Oct 23, 2026

Choose a reason for hiding this comment

Uh oh!

toddbaert Nov 3, 2026

Choose a reason for hiding this comment

Uh oh!

alexandraoberaigner Oct 23, 2026

Choose a reason for hiding this comment

Uh oh!

toddbaert Oct 23, 2026

Choose a reason for hiding this comment

Uh oh!

toddbaert Oct 23, 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!

alexandraoberaigner Oct 23, 2026

Choose a reason for hiding this comment

Uh oh!

toddbaert Oct 23, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

🟡 Changes recommended

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

🔵 Needs a closer look

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

🟢 Approval recommended

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

🟢 Approval recommended

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

🟢 Approval recommended

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

🟢 Approval recommended

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

toddbaert commented Oct 19, 2026 •

edited

Loading

toddbaert Oct 23, 2026 •

edited

Loading