blob: e76a72dcfedf138336062953d27b7870c0c52716 [file] [log] [blame]
// Copyright 2014 The Bazel Authors. All rights reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
package com.google.devtools.build.lib.actions;
import com.google.common.base.Preconditions;
import com.google.common.collect.ImmutableList;
import com.google.common.collect.ImmutableMap;
import com.google.common.collect.ImmutableSet;
import com.google.common.collect.Iterables;
import com.google.common.collect.Maps;
import com.google.common.flogger.GoogleLogger;
import com.google.devtools.build.lib.actions.Artifact.ArchivedTreeArtifact;
import com.google.devtools.build.lib.actions.Artifact.SpecialArtifact;
import com.google.devtools.build.lib.actions.cache.MetadataHandler;
import com.google.devtools.build.lib.actions.extra.ExtraActionInfo;
import com.google.devtools.build.lib.analysis.platform.PlatformInfo;
import com.google.devtools.build.lib.cmdline.Label;
import com.google.devtools.build.lib.cmdline.RepositoryMapping;
import com.google.devtools.build.lib.collect.nestedset.Depset;
import com.google.devtools.build.lib.collect.nestedset.NestedSet;
import com.google.devtools.build.lib.collect.nestedset.NestedSetBuilder;
import com.google.devtools.build.lib.collect.nestedset.Order;
import com.google.devtools.build.lib.concurrent.ThreadSafety.Immutable;
import com.google.devtools.build.lib.concurrent.ThreadSafety.ThreadSafe;
import com.google.devtools.build.lib.events.Event;
import com.google.devtools.build.lib.events.EventHandler;
import com.google.devtools.build.lib.packages.AspectDescriptor;
import com.google.devtools.build.lib.packages.semantics.BuildLanguageOptions;
import com.google.devtools.build.lib.server.FailureDetails.Execution.Code;
import com.google.devtools.build.lib.skyframe.serialization.autocodec.AutoCodec.VisibleForSerialization;
import com.google.devtools.build.lib.starlarkbuildapi.ActionApi;
import com.google.devtools.build.lib.starlarkbuildapi.CommandLineArgsApi;
import com.google.devtools.build.lib.vfs.BulkDeleter;
import com.google.devtools.build.lib.vfs.Path;
import com.google.devtools.build.lib.vfs.PathFragment;
import com.google.devtools.build.lib.vfs.Root;
import com.google.devtools.build.lib.vfs.Symlinks;
import com.google.errorprone.annotations.ForOverride;
import java.io.IOException;
import java.util.Collection;
import java.util.Map;
import javax.annotation.Nullable;
import javax.annotation.concurrent.GuardedBy;
import net.starlark.java.eval.Dict;
import net.starlark.java.eval.EvalException;
import net.starlark.java.eval.Printer;
import net.starlark.java.eval.Sequence;
import net.starlark.java.eval.StarlarkSemantics;
/**
* Abstract implementation of Action which implements basic functionality: the inputs, outputs, and
* toString method. Both input and output sets are immutable. Subclasses must be generally immutable
* - see the documentation on {@link Action}.
*/
@Immutable
@ThreadSafe
public abstract class AbstractAction extends ActionKeyCacher implements Action, ActionApi {
private static final GoogleLogger logger = GoogleLogger.forEnclosingClass();
@Override
public boolean isImmutable() {
return true; // immutable and Starlark-hashable
}
/**
* An arbitrary default resource set. We assume that a typical subprocess is single-threaded
* (i.e., uses one CPU core) and CPU-bound, and uses a small-ish amount of memory. In the past,
* we've seen that assuming less than one core can lead to local overload. Unless you have data
* indicating otherwise (for example, we've observed in the past that C++ linking can use large
* amounts of memory), we suggest to use this default set.
*/
// TODO(ulfjack): Collect actual data to confirm that this is an acceptable approximation.
public static final ResourceSet DEFAULT_RESOURCE_SET = ResourceSet.createWithRamCpu(250, 1);
/**
* The owner/inputs/outputs attributes below should never be directly accessed even within
* AbstractAction itself. The appropriate getter methods should be used instead. This has to be
* done due to the fact that the getter methods can be overridden in subclasses.
*/
@VisibleForSerialization protected final ActionOwner owner;
/**
* Tools are a subset of inputs and used by the WorkerSpawnStrategy to determine whether a
* compiler has changed since the last time it was used. This should include all artifacts that
* the tool does not dynamically reload / check on each unit of work - e.g. its own binary, the
* JDK for Java binaries, shared libraries, ... but not a configuration file, if it reloads that
* when it has changed.
*
* <p>If the "tools" set does not contain exactly the right set of artifacts, the following can
* happen: If an artifact that should be included is missing, the tool might not be restarted when
* it should, and builds can become incorrect (example: The compiler binary is not part of this
* set, then the compiler gets upgraded, but the worker strategy still reuses the old version). If
* an artifact that should *not* be included is accidentally part of this set, the worker process
* will be restarted more often that is necessary - e.g. if a file that is unique to each unit of
* work, e.g. the source code that a compiler should compile for a compile action, is part of this
* set, then the worker will never be reused and will be restarted for each unit of work.
*/
private final NestedSet<Artifact> tools;
@GuardedBy("this")
private boolean inputsDiscovered = false; // Only used when discoversInputs() returns true
// The variable inputs is non-final only so that actions that discover their inputs can modify it.
@GuardedBy("this")
@VisibleForSerialization
protected NestedSet<Artifact> inputs;
protected final ActionEnvironment env;
private final RunfilesSupplier runfilesSupplier;
@VisibleForSerialization protected final ImmutableSet<Artifact> outputs;
/** Construct an abstract action with the specified inputs and outputs; */
protected AbstractAction(
ActionOwner owner, NestedSet<Artifact> inputs, Iterable<Artifact> outputs) {
this(
owner,
/*tools=*/ NestedSetBuilder.emptySet(Order.STABLE_ORDER),
inputs,
EmptyRunfilesSupplier.INSTANCE,
outputs,
ActionEnvironment.EMPTY);
}
protected AbstractAction(
ActionOwner owner,
NestedSet<Artifact> inputs,
Iterable<Artifact> outputs,
ActionEnvironment env) {
this(
owner,
/*tools = */ NestedSetBuilder.emptySet(Order.STABLE_ORDER),
inputs,
EmptyRunfilesSupplier.INSTANCE,
outputs,
env);
}
protected AbstractAction(
ActionOwner owner,
NestedSet<Artifact> tools,
NestedSet<Artifact> inputs,
RunfilesSupplier runfilesSupplier,
Iterable<? extends Artifact> outputs,
ActionEnvironment env) {
Preconditions.checkNotNull(owner);
this.owner = owner;
this.tools = tools;
this.inputs = inputs;
this.env = Preconditions.checkNotNull(env);
this.outputs = ImmutableSet.copyOf(outputs);
this.runfilesSupplier = Preconditions.checkNotNull(runfilesSupplier);
Preconditions.checkArgument(!this.outputs.isEmpty(), "action outputs may not be empty");
}
@Override
public final ActionOwner getOwner() {
return owner;
}
@Override
public final synchronized boolean inputsDiscovered() {
return !discoversInputs() || inputsDiscovered;
}
/**
* Should be overridden by actions that do input discovery.
*
* <p>The value returned by each instance should be constant over the lifetime of that instance.
*
* <p>If this returns true, {@link #discoverInputs(ActionExecutionContext)} must also be
* implemented.
*/
@Override
public boolean discoversInputs() {
return false;
}
/**
* Runs input discovery on the action.
*
* <p>Called by Blaze if {@link #discoversInputs()} returns true. It must return the set of input
* artifacts that were not known at analysis time. May also call {@link #updateInputs}; if it
* doesn't, the action itself must arrange for the newly discovered artifacts to be available
* during action execution, probably by keeping state in the action instance and using a custom
* action execution context and for {@link #updateInputs} to be called during the execution of the
* action.
*
* <p>Since keeping state within an action is bad, don't do that unless there is a very good
* reason to do so.
*
* <p>May return {@code null} if more dependencies were requested from skyframe but were
* unavailable, meaning a restart is necessary.
*/
@Override
@Nullable
public NestedSet<Artifact> discoverInputs(ActionExecutionContext actionExecutionContext)
throws ActionExecutionException, InterruptedException {
throw new IllegalStateException("discoverInputs cannot be called for " + this.prettyPrint()
+ " since it does not discover inputs");
}
@Override
public final void resetDiscoveredInputs() {
Preconditions.checkState(discoversInputs(), "Not an input-discovering action: %s", this);
if (!inputsDiscovered()) {
return;
}
NestedSet<Artifact> originalInputs = getOriginalInputs();
if (originalInputs != null) {
synchronized (this) {
inputs = originalInputs;
inputsDiscovered = false;
}
}
}
/**
* Returns this action's <em>original</em> inputs, prior to {@linkplain #discoverInputs input
* discovery}.
*
* <p>Input-discovering actions which are able to reconstitute their original inputs may override
* this, allowing for memory savings.
*/
@Nullable
@ForOverride
protected NestedSet<Artifact> getOriginalInputs() {
return null;
}
@Override
public NestedSet<Artifact> getAllowedDerivedInputs() {
throw new IllegalStateException(
"Method must be overridden for actions that may have unknown inputs.");
}
/**
* Should be called when the inputs of the action become known, that is, either during {@link
* #discoverInputs(ActionExecutionContext)} or during {@link #execute(ActionExecutionContext)}.
*
* <p>When an action discovers inputs, it must have been called by the time {@code #execute()}
* returns. It can be called both during {@code discoverInputs} and during {@code execute()}.
*
* <p>In addition to being called from action implementations, it will also be called by Bazel
* itself when an action is loaded from the on-disk action cache.
*/
@Override
public synchronized void updateInputs(NestedSet<Artifact> inputs) {
Preconditions.checkState(
discoversInputs(), "Can't update inputs unless discovering: %s %s", this, inputs);
this.inputs = inputs;
inputsDiscovered = true;
}
@Override
public NestedSet<Artifact> getTools() {
return tools;
}
/** Should not be overridden (it's non-final only for tests) */
@Override
public synchronized NestedSet<Artifact> getInputs() {
return inputs;
}
public final ActionEnvironment getEnvironment() {
return env;
}
@Override
public ImmutableMap<String, String> getEffectiveEnvironment(Map<String, String> clientEnv)
throws CommandLineExpansionException {
Map<String, String> effectiveEnvironment = Maps.newLinkedHashMapWithExpectedSize(env.size());
env.resolve(effectiveEnvironment, clientEnv);
return ImmutableMap.copyOf(effectiveEnvironment);
}
@Override
public Collection<String> getClientEnvironmentVariables() {
return env.getInheritedEnv();
}
@Override
public RunfilesSupplier getRunfilesSupplier() {
return runfilesSupplier;
}
@Override
public ImmutableSet<Artifact> getOutputs() {
return outputs;
}
@Override
public Artifact getPrimaryInput() {
// The default behavior is to return the first input artifact.
// Call through the method, not the field, because it may be overridden.
return Iterables.getFirst(getInputs().toList(), null);
}
@Override
public Artifact getPrimaryOutput() {
// Default behavior is to return the first output artifact.
// Use the method rather than field in case of overriding in subclasses.
return Iterables.getFirst(getOutputs(), null);
}
@Override
public NestedSet<Artifact> getMandatoryInputs() {
return getInputs();
}
@Override
public String toString() {
return prettyPrint()
+ " ("
+ getMnemonic()
+ "["
+ getInputs().toList()
+ (inputsDiscovered() ? " -> " : ", unknown inputs -> ")
+ getOutputs()
+ "]"
+ ")";
}
@Override
public abstract String getMnemonic();
@Override
public String describeKey() {
return null;
}
@Override
public boolean executeUnconditionally() {
return false;
}
@Override
public boolean isVolatile() {
return false;
}
@Override
public boolean isShareable() {
return true;
}
@Override
public boolean showsOutputUnconditionally() {
return false;
}
@Nullable
@Override
public final String getProgressMessage() {
return getProgressMessageChecked(null);
}
@Nullable
@Override
public final String getProgressMessage(RepositoryMapping mainRepositoryMapping) {
Preconditions.checkNotNull(mainRepositoryMapping);
return getProgressMessageChecked(mainRepositoryMapping);
}
private String getProgressMessageChecked(@Nullable RepositoryMapping mainRepositoryMapping) {
String message = getRawProgressMessage();
if (message == null) {
return null;
}
message = replaceProgressMessagePlaceholders(message, mainRepositoryMapping);
String additionalInfo = getOwner().getAdditionalProgressInfo();
return additionalInfo == null ? message : message + " [" + additionalInfo + "]";
}
private String replaceProgressMessagePlaceholders(
String progressMessage, @Nullable RepositoryMapping mainRepositoryMapping) {
if (progressMessage.contains("%{label}") && getOwner().getLabel() != null) {
String labelString;
if (mainRepositoryMapping != null) {
labelString = getOwner().getLabel().getDisplayForm(mainRepositoryMapping);
} else {
labelString = getOwner().getLabel().toString();
}
progressMessage = progressMessage.replace("%{label}", labelString);
}
if (progressMessage.contains("%{output}") && getPrimaryOutput() != null) {
progressMessage =
progressMessage.replace("%{output}", getPrimaryOutput().getExecPathString());
}
if (progressMessage.contains("%{input}") && getPrimaryInput() != null) {
progressMessage = progressMessage.replace("%{input}", getPrimaryInput().getExecPathString());
}
return progressMessage;
}
/**
* Returns a progress message string that is specific for this action. This is then annotated with
* additional information, currently the string '[for tool]' for actions in the tool
* configurations.
*
* <p>A return value of null indicates no message should be reported.
*/
protected String getRawProgressMessage() {
// A cheesy default implementation. Subclasses are invited to do something
// more meaningful.
return defaultProgressMessage();
}
private String defaultProgressMessage() {
return getMnemonic() + " " + getPrimaryOutput().prettyPrint();
}
@Override
public String prettyPrint() {
return "action '" + describe() + "'";
}
@Override
public void repr(Printer printer) {
printer.append(prettyPrint()); // TODO(bazel-team): implement a readable representation
}
/**
* Deletes all of the action's output files, if they exist. If any of the Artifacts refers to a
* directory recursively removes the contents of the directory.
*
* @param execRoot the exec root in which this action is executed
* @param bulkDeleter a helper to bulk delete outputs to avoid delegating to the filesystem
* @param cleanupArchivedArtifacts whether to clean up archived tree artifacts
*/
protected final void deleteOutputs(
Path execRoot,
ArtifactPathResolver pathResolver,
@Nullable BulkDeleter bulkDeleter,
boolean cleanupArchivedArtifacts)
throws IOException, InterruptedException {
Iterable<Artifact> artifactsToDelete =
cleanupArchivedArtifacts
? Iterables.concat(outputs, archivedTreeArtifactOutputs())
: outputs;
Iterable<PathFragment> additionalPathOutputsToDelete = getAdditionalPathOutputsToDelete();
Iterable<PathFragment> directoryOutputsToDelete = getDirectoryOutputsToDelete();
if (bulkDeleter != null) {
bulkDeleter.bulkDelete(
Iterables.concat(
Artifact.asPathFragments(artifactsToDelete),
additionalPathOutputsToDelete,
directoryOutputsToDelete));
return;
}
// TODO(b/185277726): Either we don't need a path resolver for actual deletion of output
// artifacts (likely) or we need to transform the fragments below (and then the resolver should
// be augmented to deal with exec-path PathFragments).
for (Artifact output : artifactsToDelete) {
deleteOutput(output, pathResolver);
}
for (PathFragment path : additionalPathOutputsToDelete) {
deleteOutput(execRoot.getRelative(path), /*root=*/ null);
}
for (PathFragment path : directoryOutputsToDelete) {
execRoot.getRelative(path).deleteTree();
}
}
@ForOverride
protected Iterable<PathFragment> getAdditionalPathOutputsToDelete() {
return ImmutableList.of();
}
@ForOverride
protected Iterable<PathFragment> getDirectoryOutputsToDelete() {
return ImmutableList.of();
}
private Iterable<Artifact> archivedTreeArtifactOutputs() {
return Iterables.transform(
Iterables.filter(outputs, Artifact::isTreeArtifact),
tree -> ArchivedTreeArtifact.createForTree((SpecialArtifact) tree));
}
/**
* Remove an output artifact.
*
* <p>If the path refers to a directory, recursively removes the contents of the directory.
*
* @param output artifact to remove
*/
protected static void deleteOutput(Artifact output, ArtifactPathResolver pathResolver)
throws IOException {
deleteOutput(
pathResolver.toPath(output), pathResolver.transformRoot(output.getRoot().getRoot()));
}
/**
* Helper method to remove an output file.
*
* <p>If the path refers to a directory, recursively removes the contents of the directory.
*
* @param path the output to remove
* @param root the root containing the output. This is used to check that we don't delete
* arbitrary files in the file system.
*/
public static void deleteOutput(Path path, @Nullable Root root) throws IOException {
try {
// Optimize for the common case: output artifacts are files.
path.delete();
} catch (IOException e) {
// Handle a couple of scenarios where the output can still be deleted, but make sure we're not
// deleting random files on the filesystem.
if (root == null) {
throw new IOException("null root", e);
}
if (!root.contains(path)) {
throw new IOException(String.format("%s not under %s", path, root), e);
}
Path parentDir = path.getParentDirectory();
if (!parentDir.isWritable() && root.contains(parentDir)) {
parentDir.setWritable(true);
}
// Retry deleting after making the parent writable.
if (path.isDirectory(Symlinks.NOFOLLOW)) {
path.deleteTree();
} else {
path.delete();
}
}
}
/**
* If the action might read directories as inputs in a way that is unsound wrt dependency
* checking, this method must be called.
*/
protected void checkInputsForDirectories(
EventHandler eventHandler, MetadataProvider metadataProvider) throws ExecException {
// Report "directory dependency checking" warning only for non-generated directories (generated
// ones will be reported earlier).
for (Artifact input : getMandatoryInputs().toList()) {
// Assume that if the file did not exist, we would not have gotten here.
try {
if (input.isSourceArtifact()
&& metadataProvider.getMetadata(input).getType().isDirectory()) {
// TODO(ulfjack): What about dependency checking of special files?
eventHandler.handle(
Event.warn(
getOwner().getLocation(),
String.format(
"input '%s' to %s is a directory; "
+ "dependency checking of directories is unsound",
input.prettyPrint(), getOwner().getLabel())));
}
} catch (IOException e) {
throw new EnvironmentalExecException(e, Code.INPUT_DIRECTORY_CHECK_IO_EXCEPTION);
}
}
}
@Override
public MiddlemanType getActionType() {
return MiddlemanType.NORMAL;
}
/** If the action might create directories as outputs this method must be called. */
protected void checkOutputsForDirectories(ActionExecutionContext actionExecutionContext) {
FileArtifactValue metadata;
for (Artifact output : getOutputs()) {
MetadataHandler metadataHandler = actionExecutionContext.getMetadataHandler();
if (metadataHandler.artifactOmitted(output)) {
continue;
}
try {
metadata = metadataHandler.getMetadata(output);
} catch (IOException e) {
logger.atWarning().withCause(e).log("Error getting metadata for %s", output);
metadata = null;
}
if (metadata != null) {
if (!metadata.getType().isDirectory()) {
continue;
}
} else if (!actionExecutionContext.getInputPath(output).isDirectory()) {
continue;
}
String ownerString = Label.print(getOwner().getLabel());
actionExecutionContext
.getEventHandler()
.handle(
Event.warn(
getOwner().getLocation(),
"output '"
+ output.prettyPrint()
+ "' of "
+ ownerString
+ " is a directory; dependency checking of directories is unsound")
.withTag(ownerString));
}
}
@Override
public void prepare(
Path execRoot,
ArtifactPathResolver pathResolver,
@Nullable BulkDeleter bulkDeleter,
boolean cleanupArchivedArtifacts)
throws IOException, InterruptedException {
deleteOutputs(execRoot, pathResolver, bulkDeleter, cleanupArchivedArtifacts);
}
@Override
public final String describe() {
String progressMessage = getProgressMessage();
return progressMessage != null ? progressMessage : defaultProgressMessage();
}
@Override
public boolean shouldReportPathPrefixConflict(ActionAnalysisMetadata action) {
return this != action;
}
@Override
public ExtraActionInfo.Builder getExtraActionInfo(ActionKeyContext actionKeyContext)
throws CommandLineExpansionException, InterruptedException {
ActionOwner owner = getOwner();
ExtraActionInfo.Builder result =
ExtraActionInfo.newBuilder()
.setOwner(owner.getLabel().toString())
.setId(getKey(actionKeyContext, /*artifactExpander=*/ null))
.setMnemonic(getMnemonic());
ImmutableList<AspectDescriptor> aspectDescriptors = owner.getAspectDescriptors();
AspectDescriptor lastAspect =
aspectDescriptors.isEmpty() ? null : Iterables.getLast(aspectDescriptors);
if (lastAspect != null) {
result.setAspectName(lastAspect.getAspectClass().getName());
for (Map.Entry<String, Collection<String>> entry :
lastAspect.getParameters().getAttributes().asMap().entrySet()) {
result.putAspectParameters(
entry.getKey(),
ExtraActionInfo.StringList.newBuilder().addAllValue(entry.getValue()).build());
}
}
return result;
}
@Override
public ImmutableSet<Artifact> getMandatoryOutputs() {
return ImmutableSet.of();
}
/**
* Returns input files that need to be present to allow extra_action rules to shadow this action
* correctly when run remotely. This is at least the normal inputs of the action, but may include
* other files as well. For example C(++) compilation may perform include file header scanning.
* This needs to be mirrored by the extra_action rule. Called by {@link
* com.google.devtools.build.lib.analysis.extra.ExtraAction} at execution time for actions that
* return true for {link #discoversInputs()}.
*
* @param actionExecutionContext Services in the scope of the action, like the Out/Err streams.
* @throws ActionExecutionException only when code called from this method throws that exception.
* @throws InterruptedException if interrupted
*/
@Override
public NestedSet<Artifact> getInputFilesForExtraAction(
ActionExecutionContext actionExecutionContext)
throws ActionExecutionException, InterruptedException {
return NestedSetBuilder.emptySet(Order.STABLE_ORDER);
}
@Override
public Depset getStarlarkInputs() {
return Depset.of(Artifact.TYPE, getInputs());
}
@Override
public Depset getStarlarkOutputs() {
return Depset.of(Artifact.TYPE, NestedSetBuilder.wrap(Order.STABLE_ORDER, getOutputs()));
}
@Override
public Sequence<String> getStarlarkArgv() throws EvalException, InterruptedException {
return null;
}
@Override
public Sequence<CommandLineArgsApi> getStarlarkArgs() throws EvalException {
// Not all action types support returning Args.
return null;
}
@Override
public String getStarlarkContent() throws IOException, EvalException, InterruptedException {
return null;
}
@Override
public Dict<String, String> getStarlarkSubstitutions() throws EvalException {
return null;
}
@Override
public Dict<String, String> getExecutionInfoDict() {
ImmutableMap<String, String> executionInfo = getExecutionInfo();
return Dict.immutableCopyOf(executionInfo);
}
@Override
public Dict<String, String> getEnv(StarlarkSemantics semantics) throws EvalException {
if (semantics.getBool(BuildLanguageOptions.EXPERIMENTAL_GET_FIXED_CONFIGURED_ACTION_ENV)) {
try {
return Dict.immutableCopyOf(getEffectiveEnvironment(/*clientEnv=*/ ImmutableMap.of()));
} catch (CommandLineExpansionException ex) {
throw new EvalException(ex);
}
} else {
return Dict.immutableCopyOf(env.getFixedEnv());
}
}
@Override
public ImmutableMap<String, String> getExecProperties() {
return getOwner().getExecProperties();
}
@Override
@Nullable
public PlatformInfo getExecutionPlatform() {
return getOwner().getExecutionPlatform();
}
}