arcli write

arcli write posts a body to the write endpoint for the format you choose. The body comes from stdin or from -f FILE, the database from --database or the profile's default_database, and the token needs the write permission.

Quick reference

# Line protocol (default format)
echo "cpu,host=web-1,region=us-east usage=0.63 $(date +%s)000000000" | arcli write --database metrics
arcli write --database metrics -f cpu.lp --precision ms

# A JSON document, validated locally and sent as MessagePack
echo '{"m":"cpu","columns":{"time":[1757203380000000],"host":["web-4"],"usage":[0.39]}}' \
  | arcli write --database metrics --format json

# A MessagePack file (optionally gzip or zstd compressed), streamed unchanged
arcli write --database metrics --format msgpack -f batch.msgpack

$ arcli write --database metrics -f cpu.lp
OK
$ echo '{"m":"cpu","columns":{...}}' | arcli write --database metrics --format json
OK: 1 measurement(s), 1 row(s), 75 bytes as MessagePack

Plus the connection flags.

Line protocol

The default. Each line is measurement,tag=value field=value timestamp; the body is streamed to POST /api/v1/write/line-protocol as read, so a large file is not held in memory. --precision ns|us|ms|s tells the server the unit of the timestamps; without it the server applies its default (nanoseconds). An empty body is rejected by the server (Error: arc: Empty request body (HTTP 400)).

MessagePack

--format msgpack sends a prepared MessagePack document to POST /api/v1/write/msgpack exactly as given. Arc accepts three shapes:

• Columnar: {"m": "cpu", "columns": {"time": [...], "host": [...], "usage": [...]}}. Fastest, but every column is a field: it carries no tag metadata.
• Row: {"m": "cpu", "t": <time>, "h": "web-1", "fields": {...}, "tags": {...}}. Tags stay tags (so compaction can de-duplicate); h defaults to unknown server-side.
• Batch: {"batch": [ <row or columnar items> ]}.

A file is sent with its exact length; stdin is streamed chunked, up to the server's 1 GiB limit. Gzip and zstd bodies pass straight through, since the server sniffs the magic bytes. Timestamps carry their own unit, inferred from their magnitude, so --precision is rejected with this format. arcli checks only the first byte (a MessagePack map or array, or a compression header) and refuses a text file with a clear message.

JSON

--format json reads the same document as JSON, validates it, converts it to MessagePack in memory and posts that. The validation covers what the server would reject and what it would fail on later or drop silently: ragged or empty columns, a column mixing types, integers beyond int64, a time column mixing units, nested values, the unsupported compact f array. Integer literals become int64 (a column mixing integers and floats is promoted to float64); tags may be strings, numbers or booleans. The document is capped at 64 MiB because the conversion holds several copies; use MessagePack for bulk data. --precision is rejected here too.

Bulk loads

For files of any size, or for CSV, Parquet and TLE, use arcli import: it uploads the file to Arc's import endpoints, which parse server-side and report rows and partitions written.