CLI¶
Command-line interface for introspecting and calling methods on vgi-rpc services. Requires pip install vgi-rpc[cli].
Registered as the vgi-rpc entry point. Provides describe, call, and loggers commands.
Usage¶
Describe a service¶
# Subprocess transport
vgi-rpc describe --cmd "python worker.py"
# HTTP transport
vgi-rpc describe --url http://localhost:8000
# JSON output
vgi-rpc describe --cmd "python worker.py" --format json
Call a method¶
# Unary call with key=value parameters
vgi-rpc call add --cmd "python worker.py" a=1.0 b=2.0
# JSON parameter input
vgi-rpc call add --url http://localhost:8000 --json '{"a": 1.0, "b": 2.0}'
# Producer stream (table output)
vgi-rpc call countdown --cmd "python worker.py" n=5 --format table
# Exchange stream (pipe JSON lines to stdin)
echo '{"value": 1.0}' | vgi-rpc call accumulate --url http://localhost:8000
List loggers¶
# Human-readable table
vgi-rpc loggers
# JSON (for scripting / LLM consumption)
vgi-rpc loggers --format json
Debug logging¶
# Enable DEBUG on all vgi_rpc loggers
vgi-rpc --debug describe --cmd "python worker.py"
# Set a specific level
vgi-rpc --log-level INFO describe --cmd "python worker.py"
# Target specific loggers
vgi-rpc --log-level DEBUG --log-logger vgi_rpc.wire.request call add --cmd "python worker.py" a=1 b=2
# JSON log format on stderr
vgi-rpc --debug --log-format json call add --cmd "python worker.py" a=1 b=2
Output format¶
Results are printed as one JSON object per row (NDJSON). When a method returns a batch with multiple rows, each row is emitted as its own line rather than a single object with array values. This applies to unary results, producer streams, and exchange responses.
# A stream that emits 2 batches of 3 rows each produces 6 lines:
$ vgi-rpc call generate_rows --cmd "python worker.py" count=6 rows_per_batch=3
{"i": 0, "value": 0}
{"i": 1, "value": 10}
{"i": 2, "value": 20}
{"i": 3, "value": 30}
{"i": 4, "value": 40}
{"i": 5, "value": 50}
With --format table, rows from all batches are collected and displayed as a single aligned table:
$ vgi-rpc call generate_rows --cmd "python worker.py" count=4 --format table
i value
-- -----
0 0
1 10
2 20
3 30
With --format auto (the default), the CLI uses pretty-printed JSON when stdout is a TTY and compact NDJSON when piped.
With --format arrow, raw Arrow IPC streaming data is written to the output. Use --output/-o to direct binary output to a file:
# Unary result as Arrow IPC
vgi-rpc call add --cmd "python worker.py" a=1.0 b=2.0 --format arrow -o result.arrow
# Producer stream as Arrow IPC
vgi-rpc call generate --cmd "python worker.py" count=100 --format arrow -o data.arrow
# Stream with header: two concatenated IPC streams (header + data)
vgi-rpc call generate_with_header --cmd "python worker.py" count=5 --format arrow -o stream.arrow
The output file contains standard Arrow IPC streaming format data. For unary calls and headerless streams, this is a single IPC stream (schema + batches + EOS). For streams with headers, the file contains two concatenated IPC streams: the header stream first, then the data stream.
Read back a unary or headerless stream result:
import pyarrow as pa
from pyarrow import ipc
with ipc.open_stream(pa.OSFile("result.arrow", "rb")) as reader:
for batch in reader:
print(batch.to_pandas())
Read back a stream with header (two concatenated IPC streams):
import pyarrow as pa
from pyarrow import ipc
with open("stream.arrow", "rb") as f:
# First IPC stream: header
header_reader = ipc.open_stream(f)
header_batch = header_reader.read_next_batch()
print("Header:", header_batch.to_pydict())
# Drain remaining batches to reach EOS
for _ in header_reader:
pass
# Second IPC stream: data
data_reader = ipc.open_stream(f)
for batch in data_reader:
print(batch.to_pandas())
Without --output, arrow data is written to stdout. A warning is printed to stderr if stdout is a TTY.
Stream headers¶
Stream methods that declare a header type (e.g. Stream[MyState, MyHeader]) emit a one-time header before the data rows. The CLI reads this header and surfaces it in all output formats.
JSON (--format json): a {"__header__": {...}} line before data rows:
$ vgi-rpc call generate_with_header --cmd "python worker.py" count=3 --format json
{"__header__": {"total_count": 3, "label": "generate"}}
{"i": 0, "value": 0}
{"i": 1, "value": 10}
{"i": 2, "value": 20}
Table (--format table): a Header: section with indented key-value pairs before the data table:
$ vgi-rpc call generate_with_header --cmd "python worker.py" count=3 --format table
Header:
total_count: 3
label: generate
i value
- -----
0 0
1 10
2 20
Arrow (--format arrow): the header is written as a separate IPC stream before the data IPC stream (see reading back concatenated streams above).
Streams without headers are unaffected — no __header__ line or Header: section appears.
Options¶
| Option | Short | Description |
|---|---|---|
--url |
-u |
HTTP base URL |
--cmd |
-c |
Subprocess command |
--prefix |
-p |
URL path prefix (default /vgi) |
--format |
-f |
Output format: auto, json, table, or arrow |
--output |
-o |
Output file path (default: stdout) |
--verbose |
-v |
Show server log messages on stderr |
--debug |
Enable DEBUG on all vgi_rpc loggers to stderr |
|
--log-level |
Python logging level: DEBUG, INFO, WARNING, ERROR |
|
--log-logger |
Target specific logger(s), repeatable | |
--log-format |
Stderr log format: text (default) or json |
|
--json |
-j |
Pass parameters as a JSON string (for call) |
--debug is shorthand for --log-level DEBUG. When both are given, --debug wins. --verbose (server-to-client log messages) remains orthogonal to --debug/--log-level (Python logging).
Module Reference¶
cli
¶
Command-line interface for vgi-rpc services.
Provides describe, call, and loggers commands for introspecting
and invoking methods on any vgi-rpc service that has enable_describe=True.
Usage::
vgi-rpc describe --cmd "my-worker"
vgi-rpc call add --cmd "my-worker" a=1.0 b=2.0
vgi-rpc call generate --url http://localhost:8000 count=3
vgi-rpc loggers
OutputFormat
¶
Bases: StrEnum
Output format for CLI commands.
LogLevel
¶
Bases: StrEnum
Python logging level for --log-level.
LogFormat
¶
Bases: StrEnum
Stderr log format for --log-format.
loggers
¶
List all known vgi-rpc logger names with descriptions.
Source code in vgi_rpc/cli.py
launch
¶
launch(
ctx: Context,
socket: Annotated[
str | None,
Option(
"--socket",
help="Explicit socket path; skips hash machinery",
),
] = None,
idle_timeout: Annotated[
float,
Option(
"--idle-timeout",
help="Worker self-shutdown after N seconds idle (default 300)",
),
] = 300.0,
connect_timeout: Annotated[
float,
Option(
"--connect-timeout",
help="Max seconds to wait for the per-hash flock (default 30)",
),
] = 30.0,
worker_startup_timeout: Annotated[
float,
Option(
"--worker-startup-timeout",
help="Max seconds to wait for worker UNIX:<path> line (default 60; JVM-friendly)",
),
] = 60.0,
worker_stderr: Annotated[
str | None,
Option(
"--worker-stderr",
help="Append worker stderr to this file (default: discard)",
),
] = None,
state_dir: Annotated[
str | None,
Option(
"--state-dir",
help="Override default state directory",
),
] = None,
status: Annotated[
bool,
Option(
"--status",
help="List known workers in the state dir and exit (no spawning)",
),
] = False,
gc: Annotated[
bool,
Option(
"--gc",
help="Remove stale lock/socket/meta entries from the state dir and exit",
),
] = False,
) -> None
Discover-or-spawn a Unix-socket worker.
The actual worker command (and its args) come after -- on the command
line — typer doesn't see them; they're collected from ctx.args.
Source code in vgi_rpc/cli.py
731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 | |
describe
¶
Introspect a vgi-rpc service and show its methods.
Source code in vgi_rpc/cli.py
call
¶
call(
ctx: Context,
method: Annotated[
str, Argument(help="Method name to call")
],
args: Annotated[
list[str] | None,
Argument(help="key=value parameters"),
] = None,
json_input: Annotated[
str | None,
Option("--json", "-j", help="JSON params"),
] = None,
input_file: Annotated[
str | None,
Option(
"--input",
"-i",
help="Arrow IPC file for exchange input",
),
] = None,
no_stdin: Annotated[
bool,
Option(
"--no-stdin",
help="Force producer mode (ignore stdin)",
hidden=True,
),
] = False,
) -> None
Call a method on a vgi-rpc service.
Source code in vgi_rpc/cli.py
1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 | |