Repository navigation
Measure node availability over a trailing week, not time since registration - #36
Merged
Merged
Conversation
…ration uptime_s was the time since the node last registered with this process. Registration happens on every reconnect, config change and startup priming, and the clock was never persisted, so after a deploy every node on the leaderboard read the same few minutes. It said nothing about how reliably a node delivers. NodeMetrics now marks, in a one-bit-per-minute ring over the trailing week, each minute in which the node delivered a frame. The manager keeps the same ring for the minutes the server itself was up (the caller marks them through mark_server_up), and availability is the share of those up minutes in which the node delivered, over 7 days and over 24 hours. Minutes the server was down count for no one, so a deploy is not held against a node. A node is measured from its first whole minute after it was first seen, so a new node is not scored against a week it was not part of. A blocked node's frames still count as delivered: availability says whether a node is sending, and whether it is believed is reputation's business. NodeMetrics.to_state/from_state carry the ring, first_seen and the running counts across a restart for the caller to persist. The counts travel together because reputation reads detections per frame: detections restored without frames would read as a flood. A count missing from saved state starts at zero and a ring saved over another window is dropped, so a later change to either cannot stop a restore. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Replaces a node's
uptime_swith its availability: the share of the minutes the server was up in which the node delivered a frame, over the trailing 7 days and 24 hours.Why
uptime_swas the time since the node last registered with this process. Every reconnect, config change and startup reset it, and nothing persisted it, so after a deploy every node on the retina-server leaderboard read the same few minutes (all online nodes showed "0h 4m" on production on 2026-09-23).What changes
availability.MinuteRing: one bit per minute over the trailing week (1,260 bytes), indexed by epoch minute, so two rings compare slot for slot.NodeMetrics.record_framemarks the minute.first_seenreplacesconnected_atand is kept across re-registration.NodeAnalyticsManager.server_minutesholds the server's own up minutes, marked throughmark_server_up(). Minutes the server was down count for no one.uptime_sbecomesavailability_7d,availability_24h(shares from 0 to 1, orNoneuntil a whole minute has been measured) andavailability_measured_s.first_seen, not over the full week, so a new node is not scored against days before it existed.NodeMetrics.to_state/from_stateandMinuteRing.to_state/from_statelet the caller persist the ring,first_seenand the running counts. The counts travel together because reputation reads detections per frame. A count missing from saved state starts at zero, and a ring saved over another window is dropped, so neither can stop a restore.Tests
tests/test_availability.py, 15 tests. Each guard was checked by a control mutation that fails a test: the slot clearing as the week advances, the intersection with server minutes, the first-seen start, the minute-in-progress exclusion, re-registration keepingfirst_seen, a persisted count, the late-frame branch, the blocked path, the missing-count default and the window-length check.Full suite: 533 passed.
pre-commit run --all-filesis clean.Consumer
retina-server bumps to this commit in its own PR, which persists the state in its snapshot, runs the server clock and replaces uptime on the dashboard. ClickUp: https://app.clickup.com/t/123zgec4k7v
🤖 Generated with Claude Code