Conversation
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: e18b7c1fc0
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".
| }); | ||
| peers_mtu_guard_ = adnl::PeersMtuGuard{sender, local_id_, std::move(peers), mtu}; | ||
| // Permanent members (over validator overlays, essentially the validator set) are trusted. | ||
| peers_mtu_guard_ = adnl::PeersMtuGuard{sender, local_id_, std::move(peers), mtu, /* trusted = */ true}; |
There was a problem hiding this comment.
When an overlay configures both transports, this trusted guard is installed only on the sender selected above, which prefers twostep_broadcast_sender_. The fast-sync overlay does exactly that—legacy ADNL for two-step and QUIC for Plumtree in validator/full-node-fast-sync-overlays.cpp—so its permanent validator peers are never marked trusted in the QUIC sender; their ready QUIC connections are consequently exported as trust="untrusted". Install the permanent-peer trust registration on the Plumtree sender as well when it is distinct.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 7c7b0d8364
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".
| auto max_by_type = [&](Context family, auto value) { | ||
| family.open_family("gauge"); | ||
| for (const auto &[name, stat] : stats) { | ||
| family.with_label("type", name).with_label("window", "10m").push(value(stat)); |
There was a problem hiding this comment.
On x86 hosts whose invariant TSC is not 2 GHz, avoid labeling these samples as window="10m": ActorMetrics explicitly calibrates the real tick rate, but ActorTypeStatImpl::MaxCounter still divides ticks by the hard-coded 2 GHz Clocks::rdtsc_frequency() when rotating its 600-second buckets. For example, at 3 GHz the documented 10–20 minute retention becomes roughly 6.7–13.3 minutes, so slow actor spikes disappear from these dashboards materially earlier than operators are told; bucket rotation should use an actual time source/calibrated frequency, or the exposed window must reflect the real duration.
Useful? React with 👍 / 👎.
birydrad
changed the title
1 participant
No description provided.