DEBUG-4548 Telemetry: send events in forked children + telemetry metrics reset after fork

[p-datadog]

Worker race fix (described in https://github.com/DataDog/ruby-guild/issues/279 and originally reported as performance regression) has been extracted into #5176.

What does this PR do?

Restores #5074 and implements telemetry metrics reset after fork, which was causing wrong metric counts in datadog-ci end to end tests.

The metrics reset is done via the "at fork monkey patch". Existing worker race fix (in #5176) used the worker after-fork treatment - this PR consolidates after fork logic into the monkey patch handler.

Motivation:

Dynamic Instrumentation / Live Debugger require telemetry app-heartbeat events to properly render UI. These events are normally sent from forked children in forking web servers, and presently are missing for most customers.

Change log entry
Yes: fix Live Debugger / Dynamic Instrumentation UI for forking web servers

Additional Notes:

The original PR and the race fix are in separate commits for ease of review.

This PR now uses "at fork monkey patch" to reset the metrics manager after fork. The added tests caused existing tests to start failing due to shared global state; this was fixed in #5175.

We still need to mark the worker as "restarting after fork", but the restart is triggered from the monkey patch callback and not from enqueue call like before.

How to test the change?
New integration tests added. They have been aggravatingly flaky in CI with the previous implementation that used both worker after fork logic and at fork monkey patch, and they seem much less flaky with the current implementation that does everything from the monkey patch.

Additionally I manually tested against @anmarchenko 's reproducer.

[github-actions]

Typing analysis

Note: Ignored files are excluded from the next sections.

steep:ignore comments

This PR introduces 1 steep:ignore comment.

steep:ignore comments (+1-0)

❌ Introduced:

lib/datadog/core/telemetry/worker.rb:269

Untyped methods

This PR introduces 1 untyped method and 7 partially typed methods, and clears 1 untyped method and 7 partially typed methods. It increases the percentage of typed methods from 56.14% to 56.34% (+0.2%).

Untyped methods (+1-1)

❌ Introduced:

sig/datadog/core/telemetry/worker.rbs:61
└── def buffer_klass: () -> untyped

✅ Cleared:

sig/datadog/core/telemetry/worker.rbs:60
└── def buffer_klass: () -> untyped

Partially typed methods (+7-7)

❌ Introduced:

sig/datadog/core/telemetry/component.rbs:24
└── def self.build: (untyped settings, Datadog::Core::Configuration::AgentSettings agent_settings, Datadog::Core::Logger logger) -> Component
sig/datadog/core/telemetry/component.rbs:26
└── def initialize: (logger: Core::Logger, settings: untyped, agent_settings: Datadog::Core::Configuration::AgentSettings, enabled: true | false) -> void
sig/datadog/core/telemetry/event/app_started.rbs:19
└── def configuration: (untyped settings, Core::Configuration::AgentSettings agent_settings) -> Array[Hash[Symbol, untyped]]
sig/datadog/core/telemetry/event/app_started.rbs:23
└── def conf_value: (String name, untyped value, Integer seq_id, String origin) -> Hash[Symbol, untyped]
sig/datadog/core/telemetry/event/app_started.rbs:27
└── def install_signature: (untyped settings) -> Hash[Symbol, Object]
sig/datadog/core/telemetry/event/app_started.rbs:29
└── def get_telemetry_origin: (untyped settings, String config_path) -> String
sig/datadog/core/telemetry/event/synth_app_client_configuration_change.rbs:8
└── def payload: () -> { ?products: untyped, configuration: untyped, ?install_signature: untyped }

✅ Cleared:

sig/datadog/core/telemetry/component.rbs:23
└── def self.build: (untyped settings, Datadog::Core::Configuration::AgentSettings agent_settings, Datadog::Core::Logger logger) -> Component
sig/datadog/core/telemetry/component.rbs:25
└── def initialize: (logger: Core::Logger, settings: untyped, agent_settings: Datadog::Core::Configuration::AgentSettings, enabled: true | false) -> void
sig/datadog/core/telemetry/event/app_started.rbs:17
└── def configuration: (untyped settings, Core::Configuration::AgentSettings agent_settings) -> Array[Hash[Symbol, untyped]]
sig/datadog/core/telemetry/event/app_started.rbs:21
└── def conf_value: (String name, untyped value, Integer seq_id, String origin) -> Hash[Symbol, untyped]
sig/datadog/core/telemetry/event/app_started.rbs:25
└── def install_signature: (untyped settings) -> Hash[Symbol, Object]
sig/datadog/core/telemetry/event/app_started.rbs:27
└── def get_telemetry_origin: (untyped settings, String config_path) -> String
sig/datadog/core/telemetry/event/synth_app_client_configuration_change.rbs:8
└── def payload: () -> { configuration: untyped }

If you believe a method or an attribute is rightfully untyped or partially typed, you can add # untyped:accept to the end of the line to remove it from the stats.

[datadog-official]

✅ Tests

🎉 All green!

❄️ No new flaky tests detected
🧪 All tests passed

🎯 Code Coverage
• Patch Coverage: 82.52%
• Overall Coverage: 95.17% (-0.04%)

View detailed report

_{This comment will be updated automatically if new data arrives.
🔗 Commit SHA: 9433984 | Docs | Datadog PR Page | Was this helpful? Give us feedback!}

[pr-commenter]

Benchmarks

Benchmark execution time: 2026-01-05 19:02:02

Comparing candidate commit 9433984 in PR branch telemetry-fork-2 with baseline commit 9dcc803 in branch master.

Found 0 performance improvements and 0 performance regressions! Performance is the same for 44 metrics, 2 unstable metrics.

[p-datadog]

Worker race fix has been extracted in #5176, this PR will be rebased on top of #5176.

[p-datadog]

The tests I added are still flaking in CI. The most recent failure was on ruby 2.5 and I have seen previous failures on 2.5 also. I spent ~3 weeks trying to figure out the root cause and the problem is that once I start adding diagnostics the failures disappear. So, after 3 weeks, I am skipping the tests on 2.5 and if they fail on anything below 3.0 I will skip all of those versions also.