hydra-eval-jobs: remove, replaced by nix-eval-jobs

This commit is contained in:
Pierre Bourdon 2024-07-16 04:04:08 +02:00
parent 943aba7314
commit 273fb3434f
Signed by: delroth
GPG key ID: 6FB80DCD84DA0F1C
8 changed files with 3 additions and 650 deletions

1
.gitignore vendored
View file

@ -20,7 +20,6 @@ Makefile.in
/src/sql/hydra-postgresql.sql /src/sql/hydra-postgresql.sql
/src/sql/hydra-sqlite.sql /src/sql/hydra-sqlite.sql
/src/sql/tmp.sqlite /src/sql/tmp.sqlite
/src/hydra-eval-jobs/hydra-eval-jobs
/src/root/static/bootstrap /src/root/static/bootstrap
/src/root/static/js/flot /src/root/static/js/flot
/tests /tests

View file

@ -61,7 +61,6 @@ AC_CONFIG_FILES([
doc/manual/Makefile doc/manual/Makefile
src/Makefile src/Makefile
src/hydra-evaluator/Makefile src/hydra-evaluator/Makefile
src/hydra-eval-jobs/Makefile
src/hydra-queue-runner/Makefile src/hydra-queue-runner/Makefile
src/sql/Makefile src/sql/Makefile
src/ttf/Makefile src/ttf/Makefile

View file

@ -231,7 +231,7 @@ stdenv.mkDerivation (finalAttrs: {
shellHook = '' shellHook = ''
pushd $(git rev-parse --show-toplevel) >/dev/null pushd $(git rev-parse --show-toplevel) >/dev/null
PATH=$(pwd)/src/hydra-evaluator:$(pwd)/src/script:$(pwd)/src/hydra-eval-jobs:$(pwd)/src/hydra-queue-runner:$PATH PATH=$(pwd)/src/hydra-evaluator:$(pwd)/src/script:$(pwd)/src/hydra-queue-runner:$PATH
PERL5LIB=$(pwd)/src/lib:$PERL5LIB PERL5LIB=$(pwd)/src/lib:$PERL5LIB
export HYDRA_HOME="$(pwd)/src/" export HYDRA_HOME="$(pwd)/src/"
mkdir -p .hydra-data mkdir -p .hydra-data

View file

@ -1,3 +1,3 @@
SUBDIRS = hydra-evaluator hydra-eval-jobs hydra-queue-runner sql script lib root ttf SUBDIRS = hydra-evaluator hydra-queue-runner sql script lib root ttf
BOOTCLEAN_SUBDIRS = $(SUBDIRS) BOOTCLEAN_SUBDIRS = $(SUBDIRS)
DIST_SUBDIRS = $(SUBDIRS) DIST_SUBDIRS = $(SUBDIRS)

View file

@ -1,5 +0,0 @@
bin_PROGRAMS = hydra-eval-jobs
hydra_eval_jobs_SOURCES = hydra-eval-jobs.cc
hydra_eval_jobs_LDADD = $(NIX_LIBS) -llixcmd
hydra_eval_jobs_CXXFLAGS = $(NIX_CFLAGS) -I ../libhydra

View file

@ -1,577 +0,0 @@
#include <iostream>
#include <thread>
#include <optional>
#include <unordered_map>
#include "shared.hh"
#include "store-api.hh"
#include "eval.hh"
#include "eval-inline.hh"
#include "eval-settings.hh"
#include "signals.hh"
#include "terminal.hh"
#include "get-drvs.hh"
#include "globals.hh"
#include "lix/libcmd/common-eval-args.hh"
#include "flake/flakeref.hh"
#include "flake/flake.hh"
#include "attr-path.hh"
#include "derivations.hh"
#include "local-fs-store.hh"
#include "hydra-config.hh"
#include <sys/types.h>
#include <sys/wait.h>
#include <sys/resource.h>
#include <nlohmann/json.hpp>
void check_pid_status_nonblocking(pid_t check_pid)
{
// Only check 'initialized' and known PID's
if (check_pid <= 0) { return; }
int wstatus = 0;
pid_t pid = waitpid(check_pid, &wstatus, WNOHANG);
// -1 = failure, WNOHANG: 0 = no change
if (pid <= 0) { return; }
std::cerr << "child process (" << pid << ") ";
if (WIFEXITED(wstatus)) {
std::cerr << "exited with status=" << WEXITSTATUS(wstatus) << std::endl;
} else if (WIFSIGNALED(wstatus)) {
std::cerr << "killed by signal=" << WTERMSIG(wstatus) << std::endl;
} else if (WIFSTOPPED(wstatus)) {
std::cerr << "stopped by signal=" << WSTOPSIG(wstatus) << std::endl;
} else if (WIFCONTINUED(wstatus)) {
std::cerr << "continued" << std::endl;
}
}
using namespace nix;
static Path gcRootsDir;
static size_t maxMemorySize;
struct MyArgs : MixEvalArgs, MixCommonArgs, RootArgs
{
Path releaseExpr;
bool flake = false;
bool dryRun = false;
MyArgs() : MixCommonArgs("hydra-eval-jobs")
{
addFlag({
.longName = "gc-roots-dir",
.description = "garbage collector roots directory",
.labels = {"path"},
.handler = {&gcRootsDir}
});
addFlag({
.longName = "dry-run",
.description = "don't create store derivations",
.handler = {&dryRun, true}
});
addFlag({
.longName = "flake",
.description = "build a flake",
.handler = {&flake, true}
});
expectArg("expr", &releaseExpr);
}
};
static MyArgs myArgs;
static std::string queryMetaStrings(EvalState & state, DrvInfo & drv, const std::string & name, const std::string & subAttribute)
{
Strings res;
std::function<void(Value & v)> rec;
rec = [&](Value & v) {
state.forceValue(v, noPos);
if (v.type() == nString)
res.push_back(v.string.s);
else if (v.isList())
for (unsigned int n = 0; n < v.listSize(); ++n)
rec(*v.listElems()[n]);
else if (v.type() == nAttrs) {
auto a = v.attrs->find(state.symbols.create(subAttribute));
if (a != v.attrs->end())
res.push_back(std::string(state.forceString(*a->value, a->pos, "while evaluating meta attributes")));
}
};
Value * v = drv.queryMeta(name);
if (v) rec(*v);
return concatStringsSep(", ", res);
}
static void worker(
EvalState & state,
Bindings & autoArgs,
AutoCloseFD & to,
AutoCloseFD & from)
{
Value vTop;
if (myArgs.flake) {
using namespace flake;
auto flakeRef = parseFlakeRef(myArgs.releaseExpr);
auto vFlake = state.allocValue();
auto lockedFlake = lockFlake(state, flakeRef,
LockFlags {
.updateLockFile = false,
.useRegistries = false,
.allowUnlocked = false,
});
callFlake(state, lockedFlake, *vFlake);
auto vOutputs = vFlake->attrs->get(state.symbols.create("outputs"))->value;
state.forceValue(*vOutputs, noPos);
auto aHydraJobs = vOutputs->attrs->get(state.symbols.create("hydraJobs"));
if (!aHydraJobs)
aHydraJobs = vOutputs->attrs->get(state.symbols.create("checks"));
if (!aHydraJobs)
throw Error("flake '%s' does not provide any Hydra jobs or checks", flakeRef);
vTop = *aHydraJobs->value;
} else {
state.evalFile(lookupFileArg(state, myArgs.releaseExpr), vTop);
}
auto vRoot = state.allocValue();
state.autoCallFunction(autoArgs, vTop, *vRoot);
while (true) {
/* Wait for the master to send us a job name. */
writeLine(to.get(), "next");
auto s = readLine(from.get());
if (s == "exit") break;
if (!s.starts_with("do ")) abort();
std::string attrPath(s, 3);
debug("worker process %d at '%s'", getpid(), attrPath);
/* Evaluate it and send info back to the master. */
nlohmann::json reply;
try {
auto vTmp = findAlongAttrPath(state, attrPath, autoArgs, *vRoot).first;
auto v = state.allocValue();
state.autoCallFunction(autoArgs, *vTmp, *v);
if (auto drv = getDerivation(state, *v, false)) {
// CA derivations do not have static output paths, so we
// have to defensively not query output paths in case we
// encounter one.
DrvInfo::Outputs outputs = drv->queryOutputs(
!experimentalFeatureSettings.isEnabled(Xp::CaDerivations));
if (drv->querySystem() == "unknown")
state.error<EvalError>("derivation must have a 'system' attribute").debugThrow();
auto drvPath = state.store->printStorePath(drv->requireDrvPath());
nlohmann::json job;
job["nixName"] = drv->queryName();
job["system"] =drv->querySystem();
job["drvPath"] = drvPath;
job["description"] = drv->queryMetaString("description");
job["license"] = queryMetaStrings(state, *drv, "license", "shortName");
job["homepage"] = drv->queryMetaString("homepage");
job["maintainers"] = queryMetaStrings(state, *drv, "maintainers", "email");
job["schedulingPriority"] = drv->queryMetaInt("schedulingPriority", 100);
job["timeout"] = drv->queryMetaInt("timeout", 36000);
job["maxSilent"] = drv->queryMetaInt("maxSilent", 7200);
job["isChannel"] = drv->queryMetaBool("isHydraChannel", false);
/* If this is an aggregate, then get its constituents. */
auto a = v->attrs->get(state.symbols.create("_hydraAggregate"));
if (a && state.forceBool(*a->value, a->pos, "while evaluating the `_hydraAggregate` attribute")) {
auto a = v->attrs->get(state.symbols.create("constituents"));
if (!a)
state.error<EvalError>("derivation must have a constituents attribute").debugThrow();
NixStringContext context;
state.coerceToString(a->pos, *a->value, context, "while evaluating the `constituents` attribute", true, false);
for (auto & c : context)
std::visit(overloaded {
[&](const NixStringContextElem::Built & b) {
job["constituents"].push_back(b.drvPath->to_string(*state.store));
},
[&](const NixStringContextElem::Opaque & o) {
},
[&](const NixStringContextElem::DrvDeep & d) {
},
}, c.raw);
state.forceList(*a->value, a->pos, "while evaluating the `constituents` attribute");
for (unsigned int n = 0; n < a->value->listSize(); ++n) {
auto v = a->value->listElems()[n];
state.forceValue(*v, noPos);
if (v->type() == nString)
job["namedConstituents"].push_back(v->str());
}
}
/* Register the derivation as a GC root. !!! This
registers roots for jobs that we may have already
done. */
auto localStore = state.store.dynamic_pointer_cast<LocalFSStore>();
if (gcRootsDir != "" && localStore) {
Path root = gcRootsDir + "/" + std::string(baseNameOf(drvPath));
if (!pathExists(root))
localStore->addPermRoot(localStore->parseStorePath(drvPath), root);
}
nlohmann::json out;
for (auto & [outputName, optOutputPath] : outputs) {
if (optOutputPath) {
out[outputName] = state.store->printStorePath(*optOutputPath);
} else {
// See the `queryOutputs` call above; we should
// not encounter missing output paths otherwise.
assert(experimentalFeatureSettings.isEnabled(Xp::CaDerivations));
out[outputName] = nullptr;
}
}
job["outputs"] = std::move(out);
reply["job"] = std::move(job);
}
else if (v->type() == nAttrs) {
auto attrs = nlohmann::json::array();
StringSet ss;
for (auto & i : v->attrs->lexicographicOrder(state.symbols)) {
std::string name(state.symbols[i->name]);
if (name.find(' ') != std::string::npos) {
printError("skipping job with illegal name '%s'", name);
continue;
}
attrs.push_back(name);
}
reply["attrs"] = std::move(attrs);
}
else if (v->type() == nNull)
;
else state.error<TypeError>("attribute '%s' is %s, which is not supported", attrPath, showType(*v)).debugThrow();
} catch (EvalError & e) {
auto msg = e.msg();
// Transmits the error we got from the previous evaluation
// in the JSON output.
reply["error"] = filterANSIEscapes(msg, true);
// Don't forget to print it into the STDERR log, this is
// what's shown in the Hydra UI.
printError(msg);
}
writeLine(to.get(), reply.dump());
/* If our RSS exceeds the maximum, exit. The master will
start a new process. */
struct rusage r;
getrusage(RUSAGE_SELF, &r);
if ((size_t) r.ru_maxrss > maxMemorySize * 1024) break;
}
writeLine(to.get(), "restart");
}
int main(int argc, char * * argv)
{
/* Prevent undeclared dependencies in the evaluation via
$NIX_PATH. */
unsetenv("NIX_PATH");
return handleExceptions(argv[0], [&]() {
auto config = std::make_unique<HydraConfig>();
auto nrWorkers = config->getIntOption("evaluator_workers", 1);
maxMemorySize = config->getIntOption("evaluator_max_memory_size", 4096);
initNix();
initGC();
myArgs.parseCmdline(argvToStrings(argc, argv));
auto pureEval = config->getBoolOption("evaluator_pure_eval", myArgs.flake);
/* FIXME: The build hook in conjunction with import-from-derivation is causing "unexpected EOF" during eval */
settings.builders = "";
/* Prevent access to paths outside of the Nix search path and
to the environment. */
evalSettings.restrictEval = true;
/* When building a flake, use pure evaluation (no access to
'getEnv', 'currentSystem' etc. */
evalSettings.pureEval = pureEval;
if (myArgs.dryRun) settings.readOnlyMode = true;
if (myArgs.releaseExpr == "") throw UsageError("no expression specified");
if (gcRootsDir == "") printMsg(lvlError, "warning: `--gc-roots-dir' not specified");
struct State
{
std::set<std::string> todo{""};
std::set<std::string> active;
nlohmann::json jobs;
std::exception_ptr exc;
};
std::condition_variable wakeup;
Sync<State> state_;
/* Start a handler thread per worker process. */
auto handler = [&]()
{
Pid pid;
try {
AutoCloseFD from, to;
while (true) {
/* Start a new worker process if necessary. */
if (!pid) {
Pipe toPipe, fromPipe;
toPipe.create();
fromPipe.create();
pid = startProcess(
[&,
to{std::make_shared<AutoCloseFD>(std::move(fromPipe.writeSide))},
from{std::make_shared<AutoCloseFD>(std::move(toPipe.readSide))}
]()
{
try {
EvalState state(myArgs.searchPath, openStore());
Bindings & autoArgs = *myArgs.getAutoArgs(state);
worker(state, autoArgs, *to, *from);
} catch (Error & e) {
nlohmann::json err;
auto msg = e.msg();
err["error"] = filterANSIEscapes(msg, true);
printError(msg);
writeLine(to->get(), err.dump());
// Don't forget to print it into the STDERR log, this is
// what's shown in the Hydra UI.
writeLine(to->get(), "restart");
}
});
from = std::move(fromPipe.readSide);
to = std::move(toPipe.writeSide);
debug("created worker process %d", pid.get());
}
/* Check whether the existing worker process is still there. */
auto s = readLine(from.get());
if (s == "restart") {
pid.wait();
continue;
} else if (s != "next") {
auto json = nlohmann::json::parse(s);
throw Error("worker error: %s", (std::string) json["error"]);
}
/* Wait for a job name to become available. */
std::string attrPath;
while (true) {
checkInterrupt();
auto state(state_.lock());
if ((state->todo.empty() && state->active.empty()) || state->exc) {
writeLine(to.get(), "exit");
return;
}
if (!state->todo.empty()) {
attrPath = *state->todo.begin();
state->todo.erase(state->todo.begin());
state->active.insert(attrPath);
break;
} else
state.wait(wakeup);
}
/* Tell the worker to evaluate it. */
writeLine(to.get(), "do " + attrPath);
/* Wait for the response. */
auto response = nlohmann::json::parse(readLine(from.get()));
/* Handle the response. */
StringSet newAttrs;
if (response.find("job") != response.end()) {
auto state(state_.lock());
state->jobs[attrPath] = response["job"];
}
if (response.find("attrs") != response.end()) {
for (auto & i : response["attrs"]) {
std::string path = i;
if (path.find(".") != std::string::npos){
path = "\"" + path + "\"";
}
auto s = (attrPath.empty() ? "" : attrPath + ".") + (std::string) path;
newAttrs.insert(s);
}
}
if (response.find("error") != response.end()) {
auto state(state_.lock());
state->jobs[attrPath]["error"] = response["error"];
}
/* Add newly discovered job names to the queue. */
{
auto state(state_.lock());
state->active.erase(attrPath);
for (auto & s : newAttrs)
state->todo.insert(s);
wakeup.notify_all();
}
}
} catch (...) {
check_pid_status_nonblocking(pid.release());
auto state(state_.lock());
state->exc = std::current_exception();
wakeup.notify_all();
}
};
std::vector<std::thread> threads;
for (size_t i = 0; i < nrWorkers; i++)
threads.emplace_back(std::thread(handler));
for (auto & thread : threads)
thread.join();
auto state(state_.lock());
if (state->exc)
std::rethrow_exception(state->exc);
/* For aggregate jobs that have named consistuents
(i.e. constituents that are a job name rather than a
derivation), look up the referenced job and add it to the
dependencies of the aggregate derivation. */
auto store = openStore();
for (auto i = state->jobs.begin(); i != state->jobs.end(); ++i) {
auto jobName = i.key();
auto & job = i.value();
auto named = job.find("namedConstituents");
if (named == job.end()) continue;
std::unordered_map<std::string, std::string> brokenJobs;
auto getNonBrokenJobOrRecordError = [&brokenJobs, &jobName, &state](
const std::string & childJobName) -> std::optional<nlohmann::json> {
auto childJob = state->jobs.find(childJobName);
if (childJob == state->jobs.end()) {
printError("aggregate job '%s' references non-existent job '%s'", jobName, childJobName);
brokenJobs[childJobName] = "does not exist";
return std::nullopt;
}
if (childJob->find("error") != childJob->end()) {
std::string error = (*childJob)["error"];
printError("aggregate job '%s' references broken job '%s': %s", jobName, childJobName, error);
brokenJobs[childJobName] = error;
return std::nullopt;
}
return *childJob;
};
if (myArgs.dryRun) {
for (std::string jobName2 : *named) {
auto job2 = getNonBrokenJobOrRecordError(jobName2);
if (!job2) {
continue;
}
std::string drvPath2 = (*job2)["drvPath"];
job["constituents"].push_back(drvPath2);
}
} else {
auto drvPath = store->parseStorePath((std::string) job["drvPath"]);
auto drv = store->readDerivation(drvPath);
for (std::string jobName2 : *named) {
auto job2 = getNonBrokenJobOrRecordError(jobName2);
if (!job2) {
continue;
}
auto drvPath2 = store->parseStorePath((std::string) (*job2)["drvPath"]);
auto drv2 = store->readDerivation(drvPath2);
job["constituents"].push_back(store->printStorePath(drvPath2));
drv.inputDrvs.map[drvPath2].value = {drv2.outputs.begin()->first};
}
if (brokenJobs.empty()) {
std::string drvName(drvPath.name());
assert(drvName.ends_with(drvExtension));
drvName.resize(drvName.size() - drvExtension.size());
auto hashModulo = hashDerivationModulo(*store, drv, true);
if (hashModulo.kind != DrvHash::Kind::Regular) continue;
auto h = hashModulo.hashes.find("out");
if (h == hashModulo.hashes.end()) continue;
auto outPath = store->makeOutputPath("out", h->second, drvName);
drv.env["out"] = store->printStorePath(outPath);
drv.outputs.insert_or_assign("out", DerivationOutput::InputAddressed { .path = outPath });
auto newDrvPath = store->printStorePath(writeDerivation(*store, drv));
debug("rewrote aggregate derivation %s -> %s", store->printStorePath(drvPath), newDrvPath);
job["drvPath"] = newDrvPath;
job["outputs"]["out"] = store->printStorePath(outPath);
}
}
job.erase("namedConstituents");
/* Register the derivation as a GC root. !!! This
registers roots for jobs that we may have already
done. */
auto localStore = store.dynamic_pointer_cast<LocalFSStore>();
if (gcRootsDir != "" && localStore) {
auto drvPath = job["drvPath"].get<std::string>();
Path root = gcRootsDir + "/" + std::string(baseNameOf(drvPath));
if (!pathExists(root))
localStore->addPermRoot(localStore->parseStorePath(drvPath), root);
}
if (!brokenJobs.empty()) {
std::stringstream ss;
for (const auto& [jobName, error] : brokenJobs) {
ss << jobName << ": " << error << "\n";
}
job["error"] = ss.str();
}
}
std::cout << state->jobs.dump(2) << "\n";
});
}

View file

@ -13,7 +13,7 @@ TESTS_ENVIRONMENT = \
PGHOST=/tmp \ PGHOST=/tmp \
PERL5LIB="$(srcdir):$(abs_top_srcdir)/src/lib:$$PERL5LIB" \ PERL5LIB="$(srcdir):$(abs_top_srcdir)/src/lib:$$PERL5LIB" \
PYTHONPATH= \ PYTHONPATH= \
PATH=$(abs_top_srcdir)/src/hydra-evaluator:$(abs_top_srcdir)/src/script:$(abs_top_srcdir)/src/hydra-eval-jobs:$(abs_top_srcdir)/src/hydra-queue-runner:$$PATH \ PATH=$(abs_top_srcdir)/src/hydra-evaluator:$(abs_top_srcdir)/src/script:$(abs_top_srcdir)/src/hydra-queue-runner:$$PATH \
perl -w perl -w
EXTRA_DIST = \ EXTRA_DIST = \

View file

@ -1,63 +0,0 @@
use strict;
use warnings;
use Setup;
use Test2::V0;
use Hydra::Helper::Exec;
# Ensure that `systemd-run` is
# - Available in the PATH/envionment
# - Accessable to the user executing it
# - Capable of using the command switches we use in our test
my $sd_res;
eval {
($sd_res) = captureStdoutStderr(3, (
"systemd-run",
"--user",
"--collect",
"--scope",
"--property",
"MemoryMax=25M",
"--",
"true"
));
} or do {
# The command failed to execute, likely because `systemd-run` is not present
# in `PATH`
skip_all("`systemd-run` failed when invoked in this environment");
};
if ($sd_res != 0) {
# `systemd-run` executed but `sytemd-run` failed to call `true` and return
# successfully
skip_all("`systemd-run` returned non-zero when executing `true` (expected 0)");
}
my $ctx = test_context();
# Contain the memory usage to 25 MegaBytes using `systemd-run`
# Run `hydra-eval-jobs` on test job that will purposefully consume all memory
# available
my ($res, $stdout, $stderr) = captureStdoutStderr(60, (
"systemd-run",
"--user",
"--collect",
"--scope",
"--property",
"MemoryMax=25M",
"--",
"hydra-eval-jobs",
"-I", "/dev/zero",
"-I", $ctx->jobsdir,
($ctx->jobsdir . "/oom.nix")
));
isnt($res, 0, "`hydra-eval-jobs` exits non-zero");
ok(utf8::decode($stderr), "Stderr output is UTF8-clean");
like(
$stderr,
# Assert error log contains messages added in PR
# https://github.com/NixOS/hydra/pull/1203
qr/^child process \(\d+?\) killed by signal=9$/m,
"The stderr record includes a relevant error message"
);
done_testing;