hydra/src/hydra-queue-runner/build-remote.cc

580 lines
19 KiB
C++
Raw Normal View History

#include <algorithm>
2016-09-30 17:05:07 +02:00
#include <cmath>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include "build-result.hh"
2023-08-04 15:53:06 +02:00
#include "path.hh"
#include "serve-protocol.hh"
#include "serve-protocol-impl.hh"
2015-07-07 10:25:33 +02:00
#include "state.hh"
#include "current-process.hh"
#include "processes.hh"
2015-07-07 10:25:33 +02:00
#include "util.hh"
2023-12-04 08:56:06 -05:00
#include "serve-protocol.hh"
#include "serve-protocol-impl.hh"
#include "ssh.hh"
#include "finally.hh"
#include "url.hh"
using namespace nix;
namespace nix::build_remote {
static Strings extraStoreArgs(std::string & machine)
{
Strings result;
try {
auto parsed = parseURL(machine);
if (parsed.scheme != "ssh") {
throw SysError("Currently, only (legacy-)ssh stores are supported!");
}
machine = parsed.authority.value_or("");
auto remoteStore = parsed.query.find("remote-store");
if (remoteStore != parsed.query.end()) {
result = {"--store", shellEscape(remoteStore->second)};
}
} catch (BadURL &) {
// We just try to continue with `machine->sshName` here for backwards compat.
}
return result;
}
2024-01-22 18:38:39 -05:00
static std::unique_ptr<SSHMaster::Connection> openConnection(
2024-01-23 15:49:14 -05:00
::Machine::ptr machine, SSHMaster & master)
{
2024-01-22 18:38:39 -05:00
Strings command = {"nix-store", "--serve", "--write"};
openConnection(): Don't throw exceptions in forked child On hydra.nixos.org the queue runner had child processes that were stuck handling an exception: Thread 1 (Thread 0x7f501f7fe640 (LWP 1413473) "bld~v54h5zkhmb3"): #0 futex_wait (private=0, expected=2, futex_word=0x7f50c27969b0 <_rtld_local+2480>) at ../sysdeps/nptl/futex-internal.h:146 #1 __lll_lock_wait (futex=0x7f50c27969b0 <_rtld_local+2480>, private=0) at lowlevellock.c:52 #2 0x00007f50c21eaee4 in __GI___pthread_mutex_lock (mutex=0x7f50c27969b0 <_rtld_local+2480>) at ../nptl/pthread_mutex_lock.c:115 #3 0x00007f50c1854bef in __GI___dl_iterate_phdr (callback=0x7f50c190c020 <_Unwind_IteratePhdrCallback>, data=0x7f501f7fb040) at dl-iteratephdr.c:40 #4 0x00007f50c190d2d1 in _Unwind_Find_FDE () from /nix/store/65hafbsx91127farbmyyv4r5ifgjdg43-glibc-2.33-117/lib/libgcc_s.so.1 #5 0x00007f50c19099b3 in uw_frame_state_for () from /nix/store/65hafbsx91127farbmyyv4r5ifgjdg43-glibc-2.33-117/lib/libgcc_s.so.1 #6 0x00007f50c190ab90 in uw_init_context_1 () from /nix/store/65hafbsx91127farbmyyv4r5ifgjdg43-glibc-2.33-117/lib/libgcc_s.so.1 #7 0x00007f50c190b08e in _Unwind_RaiseException () from /nix/store/65hafbsx91127farbmyyv4r5ifgjdg43-glibc-2.33-117/lib/libgcc_s.so.1 #8 0x00007f50c1b02ab7 in __cxa_throw () from /nix/store/dd8swlwhpdhn6bv219562vyxhi8278hs-gcc-10.3.0-lib/lib/libstdc++.so.6 #9 0x00007f50c1d01abe in nix::parseURL (url="root@cb893012.packethost.net") at src/libutil/url.cc:53 #10 0x0000000000484f55 in extraStoreArgs (machine="root@cb893012.packethost.net") at build-remote.cc:35 #11 operator() (__closure=0x7f4fe9fe0420) at build-remote.cc:79 ... Maybe the fork happened while another thread was holding some global stack unwinding lock (https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71744). Anyway, since the hanging child inherits all file descriptors to SSH clients, shutting down remote builds (via 'child.to = -1' in State::buildRemote()) doesn't work and 'child.pid.wait()' hangs forever. So let's not do any significant work between fork and exec.
2022-03-30 22:39:48 +02:00
if (machine->isLocalhost()) {
2024-01-22 18:38:39 -05:00
command.push_back("--builders");
command.push_back("");
openConnection(): Don't throw exceptions in forked child On hydra.nixos.org the queue runner had child processes that were stuck handling an exception: Thread 1 (Thread 0x7f501f7fe640 (LWP 1413473) "bld~v54h5zkhmb3"): #0 futex_wait (private=0, expected=2, futex_word=0x7f50c27969b0 <_rtld_local+2480>) at ../sysdeps/nptl/futex-internal.h:146 #1 __lll_lock_wait (futex=0x7f50c27969b0 <_rtld_local+2480>, private=0) at lowlevellock.c:52 #2 0x00007f50c21eaee4 in __GI___pthread_mutex_lock (mutex=0x7f50c27969b0 <_rtld_local+2480>) at ../nptl/pthread_mutex_lock.c:115 #3 0x00007f50c1854bef in __GI___dl_iterate_phdr (callback=0x7f50c190c020 <_Unwind_IteratePhdrCallback>, data=0x7f501f7fb040) at dl-iteratephdr.c:40 #4 0x00007f50c190d2d1 in _Unwind_Find_FDE () from /nix/store/65hafbsx91127farbmyyv4r5ifgjdg43-glibc-2.33-117/lib/libgcc_s.so.1 #5 0x00007f50c19099b3 in uw_frame_state_for () from /nix/store/65hafbsx91127farbmyyv4r5ifgjdg43-glibc-2.33-117/lib/libgcc_s.so.1 #6 0x00007f50c190ab90 in uw_init_context_1 () from /nix/store/65hafbsx91127farbmyyv4r5ifgjdg43-glibc-2.33-117/lib/libgcc_s.so.1 #7 0x00007f50c190b08e in _Unwind_RaiseException () from /nix/store/65hafbsx91127farbmyyv4r5ifgjdg43-glibc-2.33-117/lib/libgcc_s.so.1 #8 0x00007f50c1b02ab7 in __cxa_throw () from /nix/store/dd8swlwhpdhn6bv219562vyxhi8278hs-gcc-10.3.0-lib/lib/libstdc++.so.6 #9 0x00007f50c1d01abe in nix::parseURL (url="root@cb893012.packethost.net") at src/libutil/url.cc:53 #10 0x0000000000484f55 in extraStoreArgs (machine="root@cb893012.packethost.net") at build-remote.cc:35 #11 operator() (__closure=0x7f4fe9fe0420) at build-remote.cc:79 ... Maybe the fork happened while another thread was holding some global stack unwinding lock (https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71744). Anyway, since the hanging child inherits all file descriptors to SSH clients, shutting down remote builds (via 'child.to = -1' in State::buildRemote()) doesn't work and 'child.pid.wait()' hangs forever. So let's not do any significant work between fork and exec.
2022-03-30 22:39:48 +02:00
} else {
2024-01-22 18:38:39 -05:00
command.splice(command.end(), extraStoreArgs(machine->sshName));
openConnection(): Don't throw exceptions in forked child On hydra.nixos.org the queue runner had child processes that were stuck handling an exception: Thread 1 (Thread 0x7f501f7fe640 (LWP 1413473) "bld~v54h5zkhmb3"): #0 futex_wait (private=0, expected=2, futex_word=0x7f50c27969b0 <_rtld_local+2480>) at ../sysdeps/nptl/futex-internal.h:146 #1 __lll_lock_wait (futex=0x7f50c27969b0 <_rtld_local+2480>, private=0) at lowlevellock.c:52 #2 0x00007f50c21eaee4 in __GI___pthread_mutex_lock (mutex=0x7f50c27969b0 <_rtld_local+2480>) at ../nptl/pthread_mutex_lock.c:115 #3 0x00007f50c1854bef in __GI___dl_iterate_phdr (callback=0x7f50c190c020 <_Unwind_IteratePhdrCallback>, data=0x7f501f7fb040) at dl-iteratephdr.c:40 #4 0x00007f50c190d2d1 in _Unwind_Find_FDE () from /nix/store/65hafbsx91127farbmyyv4r5ifgjdg43-glibc-2.33-117/lib/libgcc_s.so.1 #5 0x00007f50c19099b3 in uw_frame_state_for () from /nix/store/65hafbsx91127farbmyyv4r5ifgjdg43-glibc-2.33-117/lib/libgcc_s.so.1 #6 0x00007f50c190ab90 in uw_init_context_1 () from /nix/store/65hafbsx91127farbmyyv4r5ifgjdg43-glibc-2.33-117/lib/libgcc_s.so.1 #7 0x00007f50c190b08e in _Unwind_RaiseException () from /nix/store/65hafbsx91127farbmyyv4r5ifgjdg43-glibc-2.33-117/lib/libgcc_s.so.1 #8 0x00007f50c1b02ab7 in __cxa_throw () from /nix/store/dd8swlwhpdhn6bv219562vyxhi8278hs-gcc-10.3.0-lib/lib/libstdc++.so.6 #9 0x00007f50c1d01abe in nix::parseURL (url="root@cb893012.packethost.net") at src/libutil/url.cc:53 #10 0x0000000000484f55 in extraStoreArgs (machine="root@cb893012.packethost.net") at build-remote.cc:35 #11 operator() (__closure=0x7f4fe9fe0420) at build-remote.cc:79 ... Maybe the fork happened while another thread was holding some global stack unwinding lock (https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71744). Anyway, since the hanging child inherits all file descriptors to SSH clients, shutting down remote builds (via 'child.to = -1' in State::buildRemote()) doesn't work and 'child.pid.wait()' hangs forever. So let's not do any significant work between fork and exec.
2022-03-30 22:39:48 +02:00
}
2024-01-22 18:38:39 -05:00
return master.startCommand(std::move(command), {
"-a", "-oBatchMode=yes", "-oConnectTimeout=60", "-oTCPKeepAlive=yes"
});
}
static void copyClosureTo(
::Machine::Connection & conn,
Store & destStore,
const StorePathSet & paths,
SubstituteFlag useSubstitutes = NoSubstitute)
{
StorePathSet closure;
destStore.computeFSClosure(paths, closure);
/* Send the "query valid paths" command with the "lock" option
enabled. This prevents a race where the remote host
garbage-collect paths that are already there. Optionally, ask
the remote host to substitute missing paths. */
2016-02-26 21:15:05 +01:00
// FIXME: substitute output pollutes our build log
/* Get back the set of paths that are already valid on the remote
host. */
auto present = conn.queryValidPaths(
destStore, true, closure, useSubstitutes);
2015-06-09 16:03:41 +02:00
if (present.size() == closure.size()) return;
auto sorted = destStore.topoSortPaths(closure);
2015-06-09 16:03:41 +02:00
StorePathSet missing;
2015-06-09 16:03:41 +02:00
for (auto i = sorted.rbegin(); i != sorted.rend(); ++i)
2020-06-23 13:43:54 +02:00
if (!present.count(*i)) missing.insert(*i);
2020-06-23 13:43:54 +02:00
printMsg(lvlDebug, "sending %d missing paths", missing.size());
std::unique_lock<std::timed_mutex> sendLock(conn.machine->state->sendLock,
std::chrono::seconds(600));
conn.to << ServeProto::Command::ImportPaths;
destStore.exportPaths(missing, conn.to);
conn.to.flush();
2015-06-09 16:03:41 +02:00
if (readInt(conn.from) != 1)
2015-06-09 16:03:41 +02:00
throw Error("remote machine failed to import closure");
}
// FIXME: use Store::topoSortPaths().
static StorePaths reverseTopoSortPaths(const std::map<StorePath, UnkeyedValidPathInfo> & paths)
{
StorePaths sorted;
StorePathSet visited;
std::function<void(const StorePath & path)> dfsVisit;
dfsVisit = [&](const StorePath & path) {
if (!visited.insert(path).second) return;
auto info = paths.find(path);
auto references = info == paths.end() ? StorePathSet() : info->second.references;
for (auto & i : references)
/* Don't traverse into paths that don't exist. That can
happen due to substitutes for non-existent paths. */
if (i != path && paths.count(i))
dfsVisit(i);
sorted.push_back(path);
};
for (auto & i : paths)
dfsVisit(i.first);
return sorted;
}
static std::pair<Path, AutoCloseFD> openLogFile(const std::string & logDir, const StorePath & drvPath)
{
std::string base(drvPath.to_string());
auto logFile = logDir + "/" + std::string(base, 0, 2) + "/" + std::string(base, 2);
createDirs(dirOf(logFile));
AutoCloseFD logFD = open(logFile.c_str(), O_CREAT | O_TRUNC | O_WRONLY, 0666);
if (!logFD) throw SysError("creating log file %s", logFile);
return {std::move(logFile), std::move(logFD)};
}
static BasicDerivation sendInputs(
2022-03-21 11:35:38 +01:00
State & state,
Step & step,
Store & localStore,
Store & destStore,
::Machine::Connection & conn,
2022-03-21 11:35:38 +01:00
unsigned int & overhead,
counter & nrStepsWaiting,
counter & nrStepsCopyingTo
)
{
/* Replace the input derivations by their output paths to send a
minimal closure to the builder.
`tryResolve` currently does *not* rewrite input addresses, so it
is safe to do this in all cases. (It should probably have a mode
to do that, however, but we would not use it here.)
*/
BasicDerivation basicDrv = ({
auto maybeBasicDrv = step.drv->tryResolve(destStore, &localStore);
if (!maybeBasicDrv)
throw Error(
"the derivation '%s' cant be resolved. Its probably "
"missing some outputs",
localStore.printStorePath(step.drvPath));
*maybeBasicDrv;
});
2022-03-21 11:35:38 +01:00
/* Ensure that the inputs exist in the destination store. This is
a no-op for regular stores, but for the binary cache store,
this will copy the inputs to the binary cache from the local
store. */
if (&localStore != &destStore) {
copyClosure(localStore, destStore,
step.drv->inputSrcs,
NoRepair, NoCheckSigs, NoSubstitute);
2022-03-21 11:35:38 +01:00
}
{
auto mc1 = std::make_shared<MaintainCount<counter>>(nrStepsWaiting);
mc1.reset();
MaintainCount<counter> mc2(nrStepsCopyingTo);
printMsg(lvlDebug, "sending closure of %s to %s",
localStore.printStorePath(step.drvPath), conn.machine->sshName);
auto now1 = std::chrono::steady_clock::now();
/* Copy the input closure. */
if (conn.machine->isLocalhost()) {
StorePathSet closure;
destStore.computeFSClosure(basicDrv.inputSrcs, closure);
2022-03-21 11:35:38 +01:00
copyPaths(destStore, localStore, closure, NoRepair, NoCheckSigs, NoSubstitute);
} else {
copyClosureTo(conn, destStore, basicDrv.inputSrcs, Substitute);
2022-03-21 11:35:38 +01:00
}
auto now2 = std::chrono::steady_clock::now();
overhead += std::chrono::duration_cast<std::chrono::milliseconds>(now2 - now1).count();
}
return basicDrv;
2022-03-21 11:35:38 +01:00
}
static BuildResult performBuild(
::Machine::Connection & conn,
2022-03-21 12:14:37 +01:00
Store & localStore,
StorePath drvPath,
const BasicDerivation & drv,
const ServeProto::BuildOptions & options,
2022-03-21 12:14:37 +01:00
counter & nrStepsBuilding
)
{
conn.putBuildDerivationRequest(localStore, drvPath, drv, options);
2022-03-21 12:14:37 +01:00
BuildResult result;
time_t startTime, stopTime;
2022-03-21 12:14:37 +01:00
startTime = time(0);
2022-03-21 12:14:37 +01:00
{
MaintainCount<counter> mc(nrStepsBuilding);
result = ServeProto::Serialise<BuildResult>::read(localStore, conn);
2022-03-21 12:14:37 +01:00
}
stopTime = time(0);
if (!result.startTime) {
// If the builder gave `startTime = 0`, use our measurements
// instead of the builder's.
//
// Note: this represents the duration of a single round, rather
// than all rounds.
result.startTime = startTime;
result.stopTime = stopTime;
2022-03-21 12:14:37 +01:00
}
// If the protocol was too old to give us `builtOutputs`, initialize
// it manually by introspecting the derivation.
if (GET_PROTOCOL_MINOR(conn.remoteVersion) < 6)
{
// If the remote is too old to handle CA derivations, we cant get this
// far anyways
assert(drv.type().hasKnownOutputPaths());
DerivationOutputsAndOptPaths drvOutputs = drv.outputsAndOptPaths(localStore);
// Since this a `BasicDerivation`, `staticOutputHashes` will not
// do any real work.
auto outputHashes = staticOutputHashes(localStore, drv);
for (auto & [outputName, output] : drvOutputs) {
auto outputPath = output.second;
// Weve just asserted that the output paths of the derivation
// were known
assert(outputPath);
auto outputHash = outputHashes.at(outputName);
auto drvOutput = DrvOutput { outputHash, outputName };
result.builtOutputs.insert_or_assign(
std::move(outputName),
Realisation { drvOutput, *outputPath });
}
2022-03-21 12:14:37 +01:00
}
return result;
}
static void copyPathFromRemote(
::Machine::Connection & conn,
NarMemberDatas & narMembers,
Store & localStore,
Store & destStore,
const ValidPathInfo & info
)
{
/* Receive the NAR from the remote and add it to the
destination store. Meanwhile, extract all the info from the
NAR that getBuildOutput() needs. */
auto source2 = sinkToSource([&](Sink & sink)
{
/* Note: we should only send the command to dump the store
path to the remote if the NAR is actually going to get read
by the destination store, which won't happen if this path
is already valid on the destination store. Since this
lambda function only gets executed if someone tries to read
from source2, we will send the command from here rather
than outside the lambda. */
conn.to << ServeProto::Command::DumpStorePath << localStore.printStorePath(info.path);
conn.to.flush();
TeeSource tee(conn.from, sink);
extractNarData(tee, localStore.printStorePath(info.path), narMembers);
});
destStore.addToStore(info, *source2, NoRepair, NoCheckSigs);
}
static void copyPathsFromRemote(
::Machine::Connection & conn,
NarMemberDatas & narMembers,
Store & localStore,
Store & destStore,
const std::map<StorePath, UnkeyedValidPathInfo> & infos
)
{
auto pathsSorted = reverseTopoSortPaths(infos);
for (auto & path : pathsSorted) {
auto & info = infos.find(path)->second;
copyPathFromRemote(
conn, narMembers, localStore, destStore,
ValidPathInfo { path, info });
}
}
}
/* using namespace nix::build_remote; */
void RemoteResult::updateWithBuildResult(const nix::BuildResult & buildResult)
{
startTime = buildResult.startTime;
stopTime = buildResult.stopTime;
timesBuilt = buildResult.timesBuilt;
errorMsg = buildResult.errorMsg;
isNonDeterministic = buildResult.isNonDeterministic;
switch ((BuildResult::Status) buildResult.status) {
case BuildResult::Built:
stepStatus = bsSuccess;
break;
case BuildResult::Substituted:
case BuildResult::AlreadyValid:
stepStatus = bsSuccess;
isCached = true;
break;
case BuildResult::PermanentFailure:
stepStatus = bsFailed;
canCache = true;
errorMsg = "";
break;
case BuildResult::InputRejected:
case BuildResult::OutputRejected:
stepStatus = bsFailed;
canCache = true;
break;
case BuildResult::TransientFailure:
stepStatus = bsFailed;
canRetry = true;
errorMsg = "";
break;
case BuildResult::TimedOut:
stepStatus = bsTimedOut;
errorMsg = "";
break;
case BuildResult::MiscFailure:
stepStatus = bsAborted;
canRetry = true;
break;
case BuildResult::LogLimitExceeded:
stepStatus = bsLogLimitExceeded;
break;
case BuildResult::NotDeterministic:
stepStatus = bsNotDeterministic;
canRetry = false;
canCache = true;
break;
default:
stepStatus = bsAborted;
break;
}
}
void State::buildRemote(ref<Store> destStore,
::Machine::ptr machine, Step::ptr step,
const ServeProto::BuildOptions & buildOptions,
RemoteResult & result, std::shared_ptr<ActiveStep> activeStep,
std::function<void(StepState)> updateStep,
NarMemberDatas & narMembers)
{
2016-10-26 15:09:16 +02:00
assert(BuildResult::TimedOut == 8);
auto [logFile, logFD] = build_remote::openLogFile(logDir, step->drvPath);
AutoDelete logFileDel(logFile, false);
result.logFile = logFile;
try {
updateStep(ssConnecting);
2024-01-22 18:38:39 -05:00
SSHMaster master {
machine->sshName,
machine->sshKey,
machine->sshPublicHostKey,
false, // no SSH master yet
false, // no compression yet
logFD.get(),
};
// FIXME: rewrite to use Store.
2024-01-22 18:38:39 -05:00
auto child = build_remote::openConnection(machine, master);
{
auto activeStepState(activeStep->state_.lock());
if (activeStepState->cancelled) throw Error("step cancelled");
2024-01-22 18:38:39 -05:00
activeStepState->pid = child->sshPid;
}
Finally clearPid([&]() {
auto activeStepState(activeStep->state_.lock());
activeStepState->pid = -1;
/* FIXME: there is a slight race here with step
cancellation in State::processQueueChange(), which
could call kill() on this pid after we've done waitpid()
on it. With pid wrap-around, there is a tiny
possibility that we end up killing another
process. Meh. */
});
::Machine::Connection conn {
{
2024-01-22 18:38:39 -05:00
.to = child->in.get(),
.from = child->out.get(),
/* Handshake. */
.remoteVersion = 0xdadbeef, // FIXME avoid dummy initialize
},
/*.machine =*/ machine,
};
Finally updateStats([&]() {
2022-03-21 11:35:38 +01:00
bytesReceived += conn.from.read;
bytesSent += conn.to.written;
});
constexpr ServeProto::Version our_version = 0x206;
try {
conn.remoteVersion = decltype(conn)::handshake(
conn.to,
conn.from,
our_version,
machine->sshName);
} catch (EndOfFile & e) {
2024-01-22 18:38:39 -05:00
child->sshPid.wait();
2022-03-21 11:35:38 +01:00
std::string s = chomp(readFile(result.logFile));
2020-06-23 13:43:54 +02:00
throw Error("cannot connect to %1%: %2%", machine->sshName, s);
}
{
auto info(machine->state->connectInfo.lock());
info->consecutiveFailures = 0;
}
/* Gather the inputs. If the remote side is Nix <= 1.9, we have to
copy the entire closure of drvPath, as well as the required
outputs of the input derivations. On Nix > 1.9, we only need to
copy the immediate sources of the derivation and the required
outputs of the input derivations. */
updateStep(ssSendingInputs);
BasicDerivation resolvedDrv = build_remote::sendInputs(*this, *step, *localStore, *destStore, conn, result.overhead, nrStepsWaiting, nrStepsCopyingTo);
logFileDel.cancel();
2015-06-19 14:51:59 +02:00
/* Truncate the log to get rid of messages about substitutions
2022-03-21 11:35:38 +01:00
etc. on the remote system. */
if (lseek(logFD.get(), SEEK_SET, 0) != 0)
throw SysError("seeking to the start of log file %s", result.logFile);
if (ftruncate(logFD.get(), 0) == -1)
throw SysError("truncating log file %s", result.logFile);
logFD = -1;
/* Do the build. */
printMsg(lvlDebug, "building %s on %s",
localStore->printStorePath(step->drvPath),
machine->sshName);
updateStep(ssBuilding);
BuildResult buildResult = build_remote::performBuild(
2022-03-21 12:14:37 +01:00
conn,
*localStore,
step->drvPath,
resolvedDrv,
2022-03-21 16:33:25 +01:00
buildOptions,
2022-03-21 12:14:37 +01:00
nrStepsBuilding
);
result.updateWithBuildResult(buildResult);
if (result.stepStatus != bsSuccess) return;
result.errorMsg = "";
/* If the path was substituted or already valid, then we didn't
get a build log. */
if (result.isCached) {
printMsg(lvlInfo, "outputs of %s substituted or already valid on %s",
localStore->printStorePath(step->drvPath), machine->sshName);
unlink(result.logFile.c_str());
result.logFile = "";
}
StorePathSet outputs;
for (auto & [_, realisation] : buildResult.builtOutputs)
outputs.insert(realisation.outPath);
/* Copy the output paths. */
if (!machine->isLocalhost() || localStore != std::shared_ptr<Store>(destStore)) {
updateStep(ssReceivingOutputs);
2017-09-14 17:22:48 +02:00
MaintainCount<counter> mc(nrStepsCopyingFrom);
auto now1 = std::chrono::steady_clock::now();
auto infos = conn.queryPathInfos(*localStore, outputs);
size_t totalNarSize = 0;
for (auto & [_, info] : infos) totalNarSize += info.narSize;
if (totalNarSize > maxOutputSize) {
result.stepStatus = bsNarSizeLimitExceeded;
return;
}
/* Copy each path. */
printMsg(lvlDebug, "copying outputs of %s from %s (%d bytes)",
localStore->printStorePath(step->drvPath), machine->sshName, totalNarSize);
build_remote::copyPathsFromRemote(conn, narMembers, *localStore, *destStore, infos);
auto now2 = std::chrono::steady_clock::now();
result.overhead += std::chrono::duration_cast<std::chrono::milliseconds>(now2 - now1).count();
}
/* Register the outputs of the newly built drv */
if (experimentalFeatureSettings.isEnabled(Xp::CaDerivations)) {
auto outputHashes = staticOutputHashes(*localStore, *step->drv);
for (auto & [outputName, realisation] : buildResult.builtOutputs) {
// Register the resolved drv output
destStore->registerDrvOutput(realisation);
// Also register the unresolved one
auto unresolvedRealisation = realisation;
unresolvedRealisation.signatures.clear();
unresolvedRealisation.id.drvHash = outputHashes.at(outputName);
destStore->registerDrvOutput(unresolvedRealisation);
}
}
/* Shut down the connection. */
2024-01-22 18:38:39 -05:00
child->in = -1;
child->sshPid.wait();
} catch (Error & e) {
/* Disable this machine until a certain period of time has
passed. This period increases on every consecutive
failure. However, don't count failures that occurred soon
after the last one (to take into account steps started in
parallel). */
auto info(machine->state->connectInfo.lock());
auto now = std::chrono::system_clock::now();
if (info->consecutiveFailures == 0 || info->lastFailure < now - std::chrono::seconds(30)) {
info->consecutiveFailures = std::min(info->consecutiveFailures + 1, (unsigned int) 4);
info->lastFailure = now;
2016-09-30 17:05:07 +02:00
int delta = retryInterval * std::pow(retryBackoff, info->consecutiveFailures - 1) + (rand() % 30);
2020-06-23 13:43:54 +02:00
printMsg(lvlInfo, "will disable machine %1% for %2%s", machine->sshName, delta);
info->disabledUntil = now + std::chrono::seconds(delta);
}
throw;
2015-06-24 13:19:16 +02:00
}
}