From bca34bd17c5a6f1cebcbedd8ea56f0e64c426c5c Mon Sep 17 00:00:00 2001 From: Peter Lobsinger Date: Sun, 13 Oct 2024 09:58:56 -0700 Subject: [PATCH] perf: report unused inputs for the tar rule (#951) * perf: report unused inputs for the tar rule The `mtree` spec passed to the `tar` rule very often selects a subset of the inputs made available through the `srcs` attribute. In many cases, these subsets do not break down cleanly along dependency-tree lines and there is no simple way just pass less content to the `tar` rule. One prominent example where this occurs is when constructing the tars for OCI image layers. For instance when [building a Python-based container image](https://github.com/bazel-contrib/rules_oci/blob/main/docs/python.md), we might want to split the Python interpreter, third-party dependencies, and application code into their own layers. This is done by [filtering the `mtree_spec`](https://github.com/aspect-build/bazel-examples/blob/85cb2aaf8c6e51d5e9e086cc94b94ab896903fb0/oci_python_image/py_layer.bzl#L39). However, in the operation to construct a `tar` from a subsetted mtree, it is usually still an unsubsetted tree of `srcs` that gets passed. As a result, the subset tarball is considered dependent upon a larger set of sources than is strictly necessary. This over-scoping runs counter to a very common objective associated with breaking up an image into layers - isolating churn to a smaller slice of the application. Because of the spurious relationships established in Bazel's dependency graph, all tars get rebuilt anytime any content in the application gets changed. Tar rebuilds can even be triggered by changes to files that are completely filtered-out from all layers of the container. Redundent creation of archive content is usually not too computationally intensive, but the archives can be quite large in some cases, and avoiding a rebuild might free up gigabytes of disk and/or network bandwidth for better use. In addition, eliminating the spurious dependency edges removes erroneous constraints applied to the build action schedule; these tend to push all Tar-building operations towards the end of a build, even when some archive construction could be scheduled much earlier. ## Risk assessment and mitigation The `unused_inputs_list` mechanism used to report spurious dependency relationships is a bit difficult to use. Reporting an actually-used input as unused can create difficult to diagnose problems down the line. However, the behaviour of the `mtree`-based `tar` rule is sufficiently simple and self-contained that I am fairly confident that this rule's used/unused set can be determined accurately in a maintainable fashion. Out of an abundance of caution I have gated this feature behind a default-off flag. The `tar` rule will continue to operate as it had before - typically over-reporting dependencies - unless the `--@aspect_bazel_lib//lib:tar_compute_unused_inputs` flag is passed. ### Filter accuracy The `vis` encoding used by the `mtree` format to resiliently handle path names has a small amount of "play" to it - it is reversable but the encoded representation of a string is not unique. Two unequal encoded strings might decode to the same value; this can happen when at least one of the encoded strings contains unnecessary escapes that are nevertheless honoured by the decoder. The unused-inputs set is determined using a filter that compares `vis`-encoded strings. In the presence of non-canonically-encoded paths, false-mismatches can lead to falsely reporting that an input is unused. The only `vis`-encoded path content that is under the control of callers is the `mtree` content itself; all other `vis`-encoded strings are constructed internally to this package, not exposed publicly, and are all derived using the `lib/private/tar.bzl%_vis_encode` function; all of these paths are expected to compare exactly. Additionally, it is expected that many/most users will use this package's helpers (e.g. `mtree_spec`) when crafting their mtree content; such content is also safe. It is only when the user crafts their own mtree, or modifies an mtree spec's `content=` fields' encoding in some way, that a risk of inaccurate reporting arises. The chances for this are expected to be minor since this seems like an inconvenient and not-particularly-useful thing for a user to go out of their way to do. * Also include other bsdtar toolchain files in keep set * Add tri-state attribute to control unused-inputs behaviour This control surface provides for granular control of the feature. The interface is selected to mirror the common behaviour of `stamp` attributes. * Add bzl_library level dep * Update docs * pre-commit * Add reminder to change flag default on major-version bump * Add note about how to make unused input computation exactly correct * Add a test for unused_inputs listing * Support alternate contents= form This is accepted by bsdtar/libarchive. In fact `contents=` is the only of the pair documented in `mtree(5)`; `content=` is an undocumented alternate form supported by libarchive. * Don't try to prune the unprunable Bazel's interpretation of unused_inputs_list cannot accomodate certain things in filenames. These are also likely to mess up our own line-oriented protocol in the shellscript that produces this file. Co-authored-by: Sahin Yort * Rerun docs update --------- Co-authored-by: Sahin Yort --- docs/tar.md | 3 +- lib/BUILD.bazel | 6 ++ lib/private/BUILD.bazel | 5 +- lib/private/tar.bzl | 192 ++++++++++++++++++++++++++++++++++++-- lib/tests/tar/BUILD.bazel | 54 ++++++++++- lib/tests/tar/asserts.bzl | 42 +++++++++ 6 files changed, 292 insertions(+), 10 deletions(-) diff --git a/docs/tar.md b/docs/tar.md index 16c9596..ccdda52 100644 --- a/docs/tar.md +++ b/docs/tar.md @@ -78,7 +78,7 @@ Create an mtree specification to map a directory hierarchy. See https://man.free ## tar_rule 
-tar_rule(name, srcs, out, args, compress, mode, mtree)
+tar_rule(name, srcs, out, args, compress, compute_unused_inputs, mode, mtree)
Rule that executes BSD `tar`. Most users should use the [`tar`](#tar) macro, rather than load this directly. @@ -93,6 +93,7 @@ Rule that executes BSD `tar`. Most users should use the [`tar`](#tar) macro, rat | out | Resulting tar file to write. If absent, `[name].tar` is written. | Label | optional | `None` | | args | Additional flags permitted by BSD tar; see the man page. | List of strings | optional | `[]` | | compress | Compress the archive file with a supported algorithm. | String | optional | `""` | +| compute_unused_inputs | Whether to discover and prune input files that will not contribute to the archive.

Unused inputs are discovered by comparing the set of input files in `srcs` to the set of files referenced by `mtree`. Files not used for content by the mtree specification will not be read by the `tar` tool when creating the archive and can be pruned from the input set using the `unused_inputs_list` [mechanism](https://bazel.build/contribute/codebase#input-discovery).

Benefits: pruning unused input files can reduce the amount of work the build system must perform. Pruned files are not included in the action cache key; changes to them do not invalidate the cache entry, which can lead to higher cache hit rates. Actions do not need to block on the availability of pruned inputs, which can increase the available parallelism of builds. Pruned files do not need to be transferred to remote-execution workers, which can reduce network costs.

Risks: pruning an actually-used input file can lead to unexpected, incorrect results. The comparison performed between `srcs` and `mtree` is currently inexact and may fail to handle handwritten or externally-derived mtree specifications. However, it is safe to use this feature when the lines found in `mtree` are derived from one or more `mtree_spec` rules, filtered and/or merged on whole-line basis only.

Possible values:

- `compute_unused_inputs = 1`: Always perform unused input discovery and pruning. - `compute_unused_inputs = 0`: Never discover or prune unused inputs. - `compute_unused_inputs = -1`: Discovery and pruning of unused inputs is controlled by the --[no]@aspect_bazel_lib//lib:tar_compute_unused_inputs flag. | Integer | optional | `-1` | | mode | A mode indicator from the following list, copied from the tar manpage:

- create: Create a new archive containing the specified items. - append: Like `create`, but new entries are appended to the archive. Note that this only works on uncompressed archives stored in regular files. The -f option is required. - list: List archive contents to stdout. - update: Like `append`, but new entries are added only if they have a modification date newer than the corresponding entry in the archive. Note that this only works on uncompressed archives stored in regular files. The -f option is required. - extract: Extract to disk from the archive. If a file with the same name appears more than once in the archive, each copy will be extracted, with later copies overwriting (replacing) earlier copies. | String | optional | `"create"` | | mtree | An mtree specification file | Label | required | | diff --git a/lib/BUILD.bazel b/lib/BUILD.bazel index c1d8087..44b3128 100644 --- a/lib/BUILD.bazel +++ b/lib/BUILD.bazel @@ -31,6 +31,12 @@ bool_flag( build_setting_default = True if is_bzlmod_enabled() else False, ) +# TODO(3.0): change default to True. +bool_flag( + name = "tar_compute_unused_inputs", + build_setting_default = False, +) + config_setting( name = "enable_runfiles", values = {"enable_runfiles": "true"}, diff --git a/lib/private/BUILD.bazel b/lib/private/BUILD.bazel index b415923..2e96540 100644 --- a/lib/private/BUILD.bazel +++ b/lib/private/BUILD.bazel @@ -281,7 +281,10 @@ bzl_library( name = "tar", srcs = ["tar.bzl"], visibility = ["//lib:__subpackages__"], - deps = ["@aspect_bazel_lib//lib:paths"], + deps = [ + "@aspect_bazel_lib//lib:paths", + "@bazel_skylib//rules:common_settings", + ], ) bzl_library( diff --git a/lib/private/tar.bzl b/lib/private/tar.bzl index 7e4866a..3e96ef1 100644 --- a/lib/private/tar.bzl +++ b/lib/private/tar.bzl @@ -1,5 +1,6 @@ "Implementation of tar rule" +load("@bazel_skylib//rules:common_settings.bzl", "BuildSettingInfo") load("//lib:paths.bzl", "to_repository_relative_path") TAR_TOOLCHAIN_TYPE = "@aspect_bazel_lib//lib:tar_toolchain_type" @@ -84,6 +85,40 @@ _tar_attrs = { doc = "Compress the archive file with a supported algorithm.", values = _ACCEPTED_COMPRESSION_TYPES, ), + "compute_unused_inputs": attr.int( + doc = """ +Whether to discover and prune input files that will not contribute to the archive. + +Unused inputs are discovered by comparing the set of input files in `srcs` to the set +of files referenced by `mtree`. Files not used for content by the mtree specification +will not be read by the `tar` tool when creating the archive and can be pruned from the +input set using the `unused_inputs_list` +[mechanism](https://bazel.build/contribute/codebase#input-discovery). + +Benefits: pruning unused input files can reduce the amount of work the build system must +perform. Pruned files are not included in the action cache key; changes to them do not +invalidate the cache entry, which can lead to higher cache hit rates. Actions do not need +to block on the availability of pruned inputs, which can increase the available +parallelism of builds. Pruned files do not need to be transferred to remote-execution +workers, which can reduce network costs. + +Risks: pruning an actually-used input file can lead to unexpected, incorrect results. The +comparison performed between `srcs` and `mtree` is currently inexact and may fail to +handle handwritten or externally-derived mtree specifications. However, it is safe to use +this feature when the lines found in `mtree` are derived from one or more `mtree_spec` +rules, filtered and/or merged on whole-line basis only. + +Possible values: + + - `compute_unused_inputs = 1`: Always perform unused input discovery and pruning. + - `compute_unused_inputs = 0`: Never discover or prune unused inputs. + - `compute_unused_inputs = -1`: Discovery and pruning of unused inputs is controlled by the + --[no]@aspect_bazel_lib//lib:tar_compute_unused_inputs flag. + """, + default = -1, + values = [-1, 0, 1], + ), + "_compute_unused_inputs_flag": attr.label(default = Label("//lib:tar_compute_unused_inputs")), } _mtree_attrs = { @@ -125,6 +160,123 @@ def _calculate_runfiles_dir(default_info): return manifest.short_path[:-9] fail("manifest path {} seems malformed".format(manifest.short_path)) +def _is_unused_inputs_enabled(attr): + """Determine whether or not to compute unused inputs. + + Args: + attr: `tar` rule ctx.attr struct. Must provide `_tar_attrs`. + + Returns: bool. Whether the unused_inputs_list file should be computed. + """ + if attr.compute_unused_inputs == 1: + return True + elif attr.compute_unused_inputs == 0: + return False + elif attr.compute_unused_inputs == -1: + return attr._compute_unused_inputs_flag[BuildSettingInfo].value + else: + fail("Unexpected `compute_unused_inputs` value: {}".format(attr.compute_unused_inputs)) + +def _is_unprunable(file): + # Some input files cannot be pruned because their name will be misinterpreted by Bazel when reading the unused_inputs_list. + # * Filenames containing newlines will be mangled in the line-oriented format of the file. + # * Filenames with leading or trailing whitespace will be mangled by the call to String.trim(). + # ref https://github.com/bazelbuild/bazel/blob/678b01a512c0b28c87067cdf5a4e0224a82716c0/src/main/java/com/google/devtools/build/lib/analysis/actions/StarlarkAction.java#L357 + p = file.path + return p[0].isspace() or p[-1].isspace() or "\n" in p or "\r" in p + +def _fmt_pruanble_inputs_line(file): + if _is_unprunable(file): + return None + + # The tar.prunable_inputs.txt file has a two columns: + # 1. vis-encoded paths of the files, used in comparison + # 2. un-vis-encoded paths of the files, used for reporting back to Bazel after filtering + path = file.path + return _vis_encode(path) + " " + path + +def _fmt_keep_inputs_line(file): + # The tar.keep_inputs.txt file has a single column of vis-encoded paths of the files to keep. + return _vis_encode(file.path) + +def _configured_unused_inputs_file(ctx, srcs, keep): + """ + Compute the unused_inputs_list, if configured. + + Args: + ctx: `tar` rule context. Must provide `_tar_attrs` and a `coreutils_toolchain_type` toolchain. + srcs: sequence or depset. The set of all input sources being provided to the `tar` rule. + keep: sequence or depset. A hardcoded set of sources to consider "used" regardless of whether or not they appear in the mtree. + + Returns: file or None. List of inputs unused by the `Tar` action. + """ + if not _is_unused_inputs_enabled(ctx.attr): + return None + + coreutils = ctx.toolchains["@aspect_bazel_lib//lib:coreutils_toolchain_type"].coreutils_info.bin + + prunable_inputs = ctx.actions.declare_file(ctx.attr.name + ".prunable_inputs.txt") + keep_inputs = ctx.actions.declare_file(ctx.attr.name + ".keep_inputs.txt") + unused_inputs = ctx.actions.declare_file(ctx.attr.name + ".unused_inputs.txt") + + ctx.actions.write( + output = prunable_inputs, + content = ctx.actions.args() + .set_param_file_format("multiline") + .add_all( + srcs, + map_each = _fmt_pruanble_inputs_line, + ), + ) + ctx.actions.write( + output = keep_inputs, + content = ctx.actions.args() + .set_param_file_format("multiline") + .add_all( + keep, + map_each = _fmt_keep_inputs_line, + ), + ) + + # Unused inputs are inputs that: + # * are in the set of PRUNABLE_INPUTS + # * are not found in any content= or contents= keyword in the MTREE + # * are not in the hardcoded KEEP_INPUTS set + # + # Comparison and filtering of PRUNABLE_INPUTS is performed in the vis-encoded representation, stored in field 1, + # before being written out in the un-vis-encoded form Bazel understands, from field 2. + # + # Note: bsdtar (libarchive) accepts both content= and contents= to identify source file: + # ref https://github.com/libarchive/libarchive/blob/a90e9d84ec147be2ef6a720955f3b315cb54bca3/libarchive/archive_read_support_format_mtree.c#L1640 + # + # TODO: Make comparison exact by converting all inputs to a canonical vis-encoded form before comparing. + # See also: https://github.com/bazel-contrib/bazel-lib/issues/794 + ctx.actions.run_shell( + outputs = [unused_inputs], + inputs = [prunable_inputs, keep_inputs, ctx.file.mtree], + tools = [coreutils], + command = ''' + "$COREUTILS" join -v 1 \\ + <("$COREUTILS" sort -u "$PRUNABLE_INPUTS") \\ + <("$COREUTILS" sort -u \\ + <(grep -o '\\bcontents\\?=\\S*' "$MTREE" | "$COREUTILS" cut -d'=' -f 2-) \\ + "$KEEP_INPUTS" \\ + ) \\ + | "$COREUTILS" cut -d' ' -f 2- \\ + > "$UNUSED_INPUTS" + ''', + env = { + "COREUTILS": coreutils.path, + "PRUNABLE_INPUTS": prunable_inputs.path, + "KEEP_INPUTS": keep_inputs.path, + "MTREE": ctx.file.mtree.path, + "UNUSED_INPUTS": unused_inputs.path, + }, + mnemonic = "UnusedTarInputs", + ) + + return unused_inputs + def _tar_impl(ctx): bsdtar = ctx.toolchains[TAR_TOOLCHAIN_TYPE] inputs = ctx.files.srcs[:] @@ -151,20 +303,43 @@ def _tar_impl(ctx): src[DefaultInfo].files_to_run.repo_mapping_manifest for src in ctx.attr.srcs ] - inputs.extend([m for m in repo_mappings if m != None]) + repo_mappings = [m for m in repo_mappings if m != None] + inputs.extend(repo_mappings) + + srcs_runfiles = [ + src[DefaultInfo].default_runfiles.files + for src in ctx.attr.srcs + ] + + unused_inputs_file = _configured_unused_inputs_file( + ctx, + srcs = depset(direct = ctx.files.srcs + repo_mappings, transitive = srcs_runfiles), + keep = depset(direct = [ctx.file.mtree, bsdtar.tarinfo.binary], transitive = [bsdtar.default.files]), + ) + if unused_inputs_file: + inputs.append(unused_inputs_file) ctx.actions.run( executable = bsdtar.tarinfo.binary, - inputs = depset(direct = inputs, transitive = [bsdtar.default.files] + [ - src[DefaultInfo].default_runfiles.files - for src in ctx.attr.srcs - ]), + inputs = depset(direct = inputs, transitive = [bsdtar.default.files] + srcs_runfiles), outputs = [out], arguments = [args], mnemonic = "Tar", + unused_inputs_list = unused_inputs_file, ) - return DefaultInfo(files = depset([out]), runfiles = ctx.runfiles([out])) + # TODO(3.0): Always return a list of providers. + default_info = DefaultInfo(files = depset([out]), runfiles = ctx.runfiles([out])) + if unused_inputs_file: + return [ + default_info, + OutputGroupInfo( + # exposed for testing + _unused_inputs_file = depset([unused_inputs_file]), + ), + ] + else: + return default_info def _mtree_line(file, type, content = None, uid = "0", gid = "0", time = "1672560000", mode = "0755"): spec = [ @@ -284,5 +459,8 @@ tar = rule( doc = "Rule that executes BSD `tar`. Most users should use the [`tar`](#tar) macro, rather than load this directly.", implementation = tar_lib.implementation, attrs = tar_lib.attrs, - toolchains = [tar_lib.toolchain_type], + toolchains = [ + tar_lib.toolchain_type, + "@aspect_bazel_lib//lib:coreutils_toolchain_type", + ], ) diff --git a/lib/tests/tar/BUILD.bazel b/lib/tests/tar/BUILD.bazel index 300f56e..d2b4996 100644 --- a/lib/tests/tar/BUILD.bazel +++ b/lib/tests/tar/BUILD.bazel @@ -3,7 +3,7 @@ load("@aspect_bazel_lib//lib:diff_test.bzl", "diff_test") load("@aspect_bazel_lib//lib:tar.bzl", "mtree_mutate", "mtree_spec", "tar") load("@aspect_bazel_lib//lib:testing.bzl", "assert_archive_contains") load("@bazel_skylib//rules:write_file.bzl", "write_file") -load(":asserts.bzl", "assert_tar_listing") +load(":asserts.bzl", "assert_tar_listing", "assert_unused_listing") # The examples below work with both source files and generated files. # Here we generate a file to use in the examples. @@ -413,3 +413,55 @@ diff_test( file1 = ":strip_prefix14", file2 = "expected14.mtree", ) + +############# +# Example 15: mtree subsetting and unused-input pruning. + +copy_directory( + name = "unused_srcdir", + src = "srcdir", + out = "unused", +) + +mtree_spec( + name = "mtree15", + srcs = [ + ":treeartifact", + ], +) + +tar( + name = "tar15", + srcs = [ + "treeartifact", + ":mtree15", # Not in output archive, but cannot be pruned. + ":unused_srcdir", + ], + out = "15.tar", + compute_unused_inputs = 1, + mtree = ":mtree15", +) + +assert_tar_listing( + name = "test_unused_inputs_ignored", + actual = ":tar15", + expected = [ + "drwxr-xr-x 0 0 0 0 Jan 1 2023 lib/", + "drwxr-xr-x 0 0 0 0 Jan 1 2023 lib/tests/", + "drwxr-xr-x 0 0 0 0 Jan 1 2023 lib/tests/tar/", + "drwxr-xr-x 0 0 0 0 Jan 1 2023 lib/tests/tar/treeartifact/", + "-rwxr-xr-x 0 0 0 0 Jan 1 2023 lib/tests/tar/treeartifact/info", + "-rwxr-xr-x 0 0 0 0 Jan 1 2023 lib/tests/tar/treeartifact/pkg", + "-rwxr-xr-x 0 0 0 1 Jan 1 2023 lib/tests/tar/treeartifact/space in name.txt", + ], +) + +assert_unused_listing( + name = "test_unused_inputs_listed", + actual = ":tar15", + expected = [ + "lib/tests/tar/unused/info", + "lib/tests/tar/unused/pkg", + "lib/tests/tar/unused/space in name.txt", + ], +) diff --git a/lib/tests/tar/asserts.bzl b/lib/tests/tar/asserts.bzl index a35e19a..2884e5f 100644 --- a/lib/tests/tar/asserts.bzl +++ b/lib/tests/tar/asserts.bzl @@ -31,3 +31,45 @@ def assert_tar_listing(name, actual, expected): file2 = expected_listing, timeout = "short", ) + +# buildifier: disable=function-docstring +def assert_unused_listing(name, actual, expected): + actual_listing = native.package_relative_label("_{}_actual_listing".format(name)) + actual_shortnames = native.package_relative_label("_{}_actual_shortnames".format(name)) + actual_shortnames_file = native.package_relative_label("_{}.actual_shortnames".format(name)) + expected_listing = native.package_relative_label("_{}_expected".format(name)) + expected_listing_file = native.package_relative_label("_{}.expected".format(name)) + + native.filegroup( + name = actual_listing.name, + output_group = "_unused_inputs_file", + srcs = [actual], + testonly = True, + ) + + # Trim platform-specific bindir prefix from unused inputs listing. E.g. + # bazel-out/darwin_arm64-fastbuild/bin/lib/tests/tar/unused/info + # -> + # lib/tests/tar/unused/info + native.genrule( + name = actual_shortnames.name, + srcs = [actual_listing], + cmd = "sed 's!^bazel-out/[^/]*/bin/!!' $< >$@", + testonly = True, + outs = [actual_shortnames_file], + ) + + write_file( + name = expected_listing.name, + testonly = True, + out = expected_listing_file, + content = expected + [""], + newline = "unix", + ) + + diff_test( + name = name, + file1 = actual_shortnames, + file2 = expected_listing, + timeout = "short", + )