apollo_infra: add tcp keepalive to remote server accepted sockets

[Itay-Tsabary-Starkware]

Mirror the client-side SO_KEEPALIVE behaviour on the server: set TCP
keepalive probes on each accepted socket via a configurable idle time
(tcp_keepalive_idle_time_ms). This lets the OS detect dead clients that
disappear without sending FIN/RST.

Co-Authored-By: Claude Sonnet 4.6 noreply@anthropic.com

---

Note

Medium Risk
Touches remote server connection setup and keepalive behavior; misconfiguration could lead to unexpected connection drops or resource retention, though changes are scoped and covered by tests.

Overview
Adds TCP-level keepalive configuration on RemoteComponentServer accepted sockets using socket2 (idle time derived from keepalive_timeout_ms * TCP_KEEPALIVE_FACTOR, plus probe interval and retry count) to help the OS detect dead peers.

Reuses the existing validate_keepalive_timeout_ms validation for RemoteServerConfig, adjusts a keepalive-related client/server test naming, and expands server eviction tests with a GOAWAY-frame detector to assert HTTP/2 PING eviction still works alongside TCP keepalive.

^{Reviewed by Cursor Bugbot for commit c87c629. Bugbot is set up for automated code reviews on this repo. Configure here.}

[reviewable-StarkWare]

This change is 

[Itay-Tsabary-Starkware]

• apollo_infra: add tcp keepalive to remote server accepted sockets #13456 👈 (View in Graphite)
• apollo_infra: move server connection eviction test to designated module #13697
• apollo_infra: add client connection eviction test #13696
• apollo_infra: add keepalive time greater than 1 validation #13694 : 1 other dependent PR (#13559 )
• apollo_infra: rename and move definition of tcp keepalive factor #13693
• apollo_infra: move test utils to relevant module #13692
• apollo_infra: organize tests #13686
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

[github-actions]

Artifacts upload workflows:

• Started at 2026-03-24T19:49:24Z
• Started at 2026-03-24T19:59:24Z
• Started at 2026-04-09T09:46:51Z

[matanl-starkware]

@matanl-starkware reviewed 19 files and all commit messages, and made 1 comment.
Reviewable status: all files reviewed, 1 unresolved discussion (waiting on Itay-Tsabary-Starkware).

---

crates/apollo_infra/src/component_server/remote_component_server.rs line 88 at r1 (raw file):

    /// TCP keepalive: milliseconds an accepted socket may be idle before the OS sends probes.
    /// `None` leaves TCP keepalive unset (OS defaults apply).
    pub tcp_keepalive_idle_time_ms: Option<u64>,

Why do we need both tcp_keepalive and hyper_keepalive?

Code quote:

pub tcp_keepalive_idle_time_ms: Option<u64>,

[Itay-Tsabary-Starkware]

@Itay-Tsabary-Starkware made 1 comment.
Reviewable status: 0 of 23 files reviewed, 1 unresolved discussion (waiting on matanl-starkware).

---

crates/apollo_infra/src/component_server/remote_component_server.rs line 88 at r1 (raw file):

Previously, matanl-starkware (Matan Lior) wrote…

Why do we need both tcp_keepalive and hyper_keepalive?

removed, per our discussion

[graphite-app]

Merge activity

• Mar 30, 9:28 AM UTC: Graphite rebased this pull request, because this pull request is set to merge when ready.
• Apr 12, 12:38 PM UTC: Graphite rebased this pull request, because this pull request is set to merge when ready.

[matanl-starkware]

@matanl-starkware reviewed 7 files and all commit messages, made 1 comment, and resolved 1 discussion.
Reviewable status: 7 of 23 files reviewed, 1 unresolved discussion (waiting on Itay-Tsabary-Starkware).

---

crates/apollo_infra/src/component_client/remote_component_client.rs line 121 at r4 (raw file):

    if tcp_keepalive >= http_keepalive {
        Ok(())
    } else {

This is guaranteed by the factor check (>1.0)

Code quote:

    if tcp_keepalive >= http_keepalive {
        Ok(())
    } else {

[matanl-starkware]

@matanl-starkware reviewed 5 files and all commit messages, made 2 comments, and resolved 1 discussion.
Reviewable status: 7 of 25 files reviewed, all discussions resolved (waiting on Itay-Tsabary-Starkware).

---

crates/apollo_infra/src/component_client/remote_component_client_test.rs line 22 at r6 (raw file):

// keepalive_timeout_ms = 1100 ms: tcp_raw = 1650 ms, tcp_whole_secs = 1 s = 1000 ms, which is
// less than http_keepalive = 1100 ms.

Just make sure this test's duration is under 300 ms.

Code quote:

1100

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, have a team admin enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 2874529. Configure here.}

[cursor]

PR Summary

Medium Risk
Touches connection-level networking behavior by enabling and validating TCP keepalive on server-accepted sockets; misconfiguration could change connection liveness/eviction timing or surface OS-specific socket option issues.

Overview
Adds server-side TCP keepalive configuration to RemoteComponentServer by applying SO_KEEPALIVE (idle time derived via TCP_KEEPALIVE_FACTOR, interval, and retry count) on each accepted TCP stream.

Reuses the client’s validate_keepalive_timeout_ms to validate RemoteServerConfig.keepalive_timeout_ms, promotes socket2 to a normal dependency, and updates/extends tests to assert keepalive options and to ensure TCP keepalive does not interfere with existing HTTP/2 PING-based zombie connection eviction (including a new contains_goaway_frame helper).

^{Reviewed by Cursor Bugbot for commit 7e7db02. Bugbot is set up for automated code reviews on this repo. Configure here.}

[Itay-Tsabary-Starkware]

@Itay-Tsabary-Starkware reviewed 26 files and all commit messages.
Reviewable status: complete! all files reviewed, all discussions resolved (waiting on Itay-Tsabary-Starkware).