Qdrant Telemetry Monitor
Real-time TUI dashboard for monitoring Qdrant telemetry. Polls the Qdrant telemetry endpoint every 500ms and displays live-updating charts.
Usage
cargo runBy default the tool connects to http://localhost:6333/telemetry?details_level=100.
CLI Flags
| Flag | Description |
|---|---|
--skip-access |
Ignore access.json and use the default localhost URL |
--queue |
Show the update queue length plot |
--search |
Show the search latency plot |
--points |
Show the segment points (plain vs indexed) plot |
--segments |
Show the segments count plot |
When no plot flags are specified, --queue, --search, and --segments are enabled by default. The --points plot is off by default.
To show only specific plots:
cargo run -- --queue --segmentsTo show all plots including points:
cargo run -- --queue --search --points --segmentsRemote Access (access.json)
To monitor a remote Qdrant instance, create an access.json file in the working directory:
{
"url": "https://your-qdrant-instance.example.com:6333",
"api_key": "your-api-key-here"
}The tool will automatically append /telemetry?details_level=100 to the URL. The status bar shows [REMOTE] or [LOCAL] to indicate which mode is active.
If no access.json is found (or --skip-access is used), the tool falls back to localhost.
Keyboard Controls
| Key | Action |
|---|---|
q / Esc |
Quit |
r |
Reset all data and start fresh |
s |
Save the last received telemetry JSON to a timestamped file (e.g. telemetry_20260210_220852.json) |
d |
Drop the first half of all collected data (useful to free memory on long runs) |
Plots
Status Bar
The top row shows the connection status and current values:
- Waiting for Qdrant process... (yellow) — no successful response yet
- Qdrant ONLINE (green) — connected; shows current queue length, search latency, and sample count
- Qdrant process OFFLINE (red) — was connected but lost connection; waiting for reconnect
Recording starts automatically once the update queue length becomes non-zero.
Update Queue Length
Tracks the number of pending updates in Qdrant's internal update queue over time. A rising queue indicates that writes are arriving faster than Qdrant can process them. When the queue drops back to zero, all updates have been applied.
- Color: Cyan
Search Latency
Split into two stacked sub-charts with independent Y-axis scales:
- QueryBatch (top, magenta) — average duration in microseconds of gRPC
/qdrant.Points/QueryBatchrequests - filtered_small_cardinality (bottom, blue) — sum of
filtered_small_cardinality.avg_duration_microsacross all segments, showing per-segment filtered search cost
Segment Points (plain vs indexed)
Shows the total number of points split by segment type:
- plain (yellow) — points in plain (non-indexed) segments
- indexed (green) — points in segments with built indexes
During indexing, you'll see points move from plain to indexed as segments get optimized.
Segments & Pending Optimizations
Tracks segment counts and optimization activity:
- total (white) — total number of segments
- plain (yellow) — segments with plain index
- non-plain (green) — segments with built indexes
- pending-opt (red) — optimization log entries that are not in "done" status
Below the chart is a color-coded optimization status bar showing what type of optimization is running at each point in time:
| Color | Meaning |
|---|---|
| Green | Idle (no active optimization) |
| Red | Merge optimization running |
| Yellow | Indexing optimization running |
| Magenta | Both merge and indexing running simultaneously |