subscribe

This document provides an overview of the commands and options available under buck2 subscribe.

buck2 subscribe

Open a subscription channel to the Buck2 daemon. This allows you to interact with the Buck2 daemon via the stdin and stdout of this command: you send requests to the daemon by writing to stdin, and you get responses via stdout.

The protocol used by this command is length-prefixed protobuf. This format is a repeated series of a varint followed by a record of the length indicated by said varint.

The protobuf spec for those records is described in buck2_subscription_proto/subscription.proto. The client writes SubscriptionRequest and reads SubscriptionResponse. See the documentation in subscription.proto to discover available APIs.

This API does not (currently) allow invalid requests and will error out when one is sent.

Usage: buck2 subscribe [OPTIONS]

Common Options:

Common options are documented on the Common Options page.

Options:

• --active-commands Whether to request command snapshots

• --unstable-json Whether to get output as JSON. The JSON format is deemed unstable so this should only be used for debugging

• --disable-starlark-types Disable runtime type checking in Starlark interpreter.

  This option is not stable, and can be used only locally to diagnose evaluation performance problems.

• --stack Record or show target call stacks.

  Starlark call stacks will be included in duplicate targets error.

  If a command outputs targets (like targets command), starlark call stacks will be printed after the targets.

• --profile-patterns <PROFILE_PATTERNS> Enables profiling for all evaluations whose evaluation identifier matches one of the provided patterns.

  Some examples identifiers: analysis/cell//buck2/app/buck2_action_impl:buck2_action_impl (cfg:linux-x86_64#27ac5723e0c99706) load/cell//build_defs/json.bzl load/prelude//playground/test.bxl load/cell//build_defs/json.bzl@other_cell load_buildfile/fbcode//third-party-buck/platform010/build/ncurses load_packagefile/fbcode//cli/rust/cli_delegate anon_analysis/anon//:_anon_link_rule (anon: 766183dc9b6f680a) (fbcode//buck2/platform/execution:linux-x86_64#08961b14cfb182aa) bxl/prelude//playground/test.bxl:playground

  You can pass --profile-patterns=.* to enable no-op profiling for everything (additionally pass --profile-patterns-mode=none to use no-op profiling to just get a list of all the identifiers).

  The profile results will be written to individual .profile files in <ROOT_OUTPUT>/<data+time>-<uuid>/ where ROOT_OUTPUT comes from the --profile-patterns-output flag. In that directory there will also be a file listing all the identifiers that were profiled.

  Enabling/disabling profiling of an evaluation will invalidate the results of that evaluation and it will be recomputed. In some cases, this will cause other work to also need to be redone (for example, invalidating the result of loading PACKAGE files causes all consumers to be recomputed). But if you keep profiling options consistent between commands, only the work that is otherwise invalidated will be redone (and only for those would profiling results be created).

  You must also pass --profile-patterns-mode and --profile-patterns-output.

• --profile-patterns-output <PATH>

• --profile-patterns-mode <PROFILE_PATTERNS_MODE> Profile mode.

  Memory profiling modes have suffixes either -allocated or -retained.

  -retained means memory kept in frozen starlark heaps after analysis completes. -retained does not work when profiling loading, because no memory is retained after loading and frozen heap is not even created. This is probably what you want when profiling analysis.

  -allocated means allocated memory, including memory which is later garbage collected.

  • Possible values:
    • time-flame
    • heap-allocated
    • heap-retained
    • heap-flame-allocated
    • heap-flame-retained
    • heap-summary-allocated
    • heap-summary-retained
    • statement
    • bytecode
    • bytecode-pairs
    • typecheck
    • coverage
    • none