Repository navigation
release: v1.4.3 — live speed (#93), domain export (#54), LOG_LEVEL (#80); agent GitHub token (#77) - #99
Conversation
…chart (#93) - Download/Upload cards show the current rate, derived from successive summary totals averaged over a 15s window. Shown only for live presets; the window restarts when the rolling range shifts and is cleared when the backend changes, so a moving window never yields negative or huge rates. Stale readings expire after 30s. - The trend chart gains a Traffic / Speed toggle; speed is bytes per bucket divided by the bucket length, with B/s units on axis and tooltip. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Codex (gpt-6-astra) <noreply@openai.com>
… retries (#77) Co-Authored-By: Codex (gpt-6-astra) <noreply@openai.com>
Co-Authored-By: Codex (gpt-6-astra) <noreply@openai.com>
The request is about domains hit by a proxy or rule, which live in the shared DomainStatsTable of the rule, proxy and device detail views. Add the export button there too; it exports the full filtered, sorted list (not just the current page) named after the rule, proxy or device. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The newest trend bucket is still filling, so dividing it by the full bucket length under-reported the current speed. Divide by the time up to the query end instead (at least 60s, since the end is minute-rounded). Speed axis ticks use one decimal and a wider axis so labels aren't clipped. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The query end is minute-rounded, so the newest bucket was still under-reported. A 5s clock (state set from a timer, not during render) gives the elapsed time of the bucket that is still filling. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The newest bucket keeps receiving bytes for a while after it starts, so any rate computed from it dips at the right edge. Speed mode now plots completed buckets only; the live rate is shown on the stat cards. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…token Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…93) Queries keep the previous data while a new window or backend loads, so the first totals after a switch belonged to the old window. Using them as the baseline turned the jump to the new totals into GB/s readings for up to 15s (review B1). The tracker is now a pure stepRateTracker() that holds back those totals until they change. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…#93) On first load the HTTP summary carries persisted totals only, while WS pushes also include the realtime buffer, so the first delta read ~4x the real speed in the browser. After a switch the tracker now also holds back the first fresh totals and starts sampling on the next change. A 5s minimum span damps push-timing jitter. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…flushed Persisted totals are cached (DB range cache 8s, WS base summary 2s) while realtime deltas are cleared right after each flush. Until those caches expired, 'cached persisted + realtime' missed a whole flush interval, so the live Total Download/Upload cards dipped by ~30s of traffic every flush and jumped back on the next push (seen while verifying #93). RealtimeStore now notifies onCleared(); the entry point drops the DB range cache and the WS base-summary cache for that backend. This covers the mihomo, Surge and agent flush paths, which all clear through the store. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Track 'has ever received data' in state (adjusted during render) instead of reading a ref during render. - Hoist the tooltip into a module-level component so it is not recreated on every render. - Widen the traffic-mode Y axis so 3-digit MB labels are not clipped. Brings trend-chart.tsx from 7 lint errors (1 on main) to 0. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Review B2: after a reconnect the dashboard briefly showed the last push from before the disconnect, while HTTP fallback had already moved the totals forward. That stale value became the rate baseline and the whole outage was spread over a few seconds (about 11x after 60s down). - websocket: clear lastMessage on close; stats pushes now carry the range they were computed for (captured before the async query) and the client drops pushes for another range. - transfer-rate: within one window totals never decrease, so a lower reading is ignored instead of becoming a new baseline. - collector: reuse clearBackendCache for the flush invalidation. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
![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: 55f353c10c
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".
| tag="$(curl -fsSL "$api_url" 2>/dev/null | awk -F'"' '/"tag_name"/{print $4}' | grep '^agent-v' | head -1)" | ||
| set -- -fsSL --connect-timeout 10 --max-time 120 --retry 3 --retry-delay 2 | ||
| if [ -n "$token" ]; then | ||
| set -- "$@" -H "Authorization: Bearer $token" |
There was a problem hiding this comment.
When GITHUB_TOKEN or GITHUB_PAT is set, the literal bearer token is expanded into the curl or wget argument list. During a slow or retried API request, another local user or process monitor can read the downloader's command line and recover the token, which may be a broadly scoped PAT. Pass the header through a protected config file or another non-argv channel; the same exposure is duplicated in install.sh.
Useful? React with 👍 / 👎.
| const scale = mode === "speed" ? bucketSeconds : 1; | ||
| return { | ||
| time: point.time, | ||
| download: point.download, | ||
| upload: point.upload, | ||
| download: point.download / scale, | ||
| upload: point.upload / scale, |
There was a problem hiding this comment.
When a query boundary is not aligned to bucketMinutes—for example, a normal 24-hour range ending at :03 with ten-minute buckets—the repository groups only the included minutes into the first boundary bucket, but this code divides that partial total by the full bucket duration. The first displayed speed is therefore systematically low, and historical ranges can similarly include a partial final bucket. Exclude both clipped boundary buckets or divide them by their actual covered duration.
Useful? React with 👍 / 👎.
| onClick={() => | ||
| downloadTextFile( | ||
| getDomainsExportFilename(ruleName || sourceChain || sourceIP || title || "domains"), | ||
| filteredDomains.map((item) => item.domain).filter(Boolean), | ||
| ) |
There was a problem hiding this comment.
For a rule, proxy, or device with more than 5,000 associated domains, filteredDomains contains only the bounded detail-query result (QUERY_CONFIG.LIMIT.DETAIL is 5,000), so this button silently exports a truncated rule set. The IP expansion path has the same issue and is capped by its controller at 2,000 domains. Paginate or use a dedicated export endpoint so the downloaded file contains the complete matching set.
Useful? React with 👍 / 👎.
Release v1.4.3 (collector/web) and agent-v1.4.8.
Added
LOG_LEVEL. Sets collector log verbosity:error,warn,info(default) ordebug.console.erroris always printed.nekoagent upgradeandinstall.shsendGITHUB_TOKENtoapi.github.comonly. Downloads and API calls now have timeouts and retries.Fixed (found while verifying #93)
Verification
next build, lint (trend-chart now 0 issues; web total 110 errors vs 111 on main)sh -nanddash -nLOG_LEVEL=errorreduced collector log lines from 24 to 0.🤖 Generated with Claude Code