lix/src/libstore/misc.cc

284 lines
8.7 KiB
C++
Raw Normal View History

#include "derivations.hh"
#include "parsed-derivations.hh"
#include "globals.hh"
#include "local-store.hh"
#include "store-api.hh"
#include "thread-pool.hh"
2020-07-27 20:45:34 +00:00
#include "topo-sort.hh"
#include "callback.hh"
namespace nix {
void Store::computeFSClosure(const StorePathSet & startPaths,
StorePathSet & paths_, bool flipDirection, bool includeOutputs, bool includeDerivers)
{
struct State
{
size_t pending;
StorePathSet & paths;
std::exception_ptr exc;
};
Sync<State> state_(State{0, paths_, 0});
std::function<void(const StorePath &)> enqueue;
std::condition_variable done;
enqueue = [&](const StorePath & path) -> void {
{
auto state(state_.lock());
if (state->exc) return;
if (!state->paths.insert(path).second) return;
state->pending++;
}
queryPathInfo(path, {[&](std::future<ref<const ValidPathInfo>> fut) {
2018-03-27 20:16:01 +00:00
// FIXME: calls to isValidPath() should be async
try {
auto info = fut.get();
if (flipDirection) {
StorePathSet referrers;
queryReferrers(path, referrers);
for (auto & ref : referrers)
if (ref != path)
enqueue(ref);
if (includeOutputs)
for (auto & i : queryValidDerivers(path))
enqueue(i);
if (includeDerivers && path.isDerivation())
for (auto & i : queryDerivationOutputs(path))
if (isValidPath(i) && queryPathInfo(i)->deriver == path)
enqueue(i);
} else {
for (auto & ref : info->references)
if (ref != path)
enqueue(ref);
if (includeOutputs && path.isDerivation())
for (auto & i : queryDerivationOutputs(path))
if (isValidPath(i)) enqueue(i);
if (includeDerivers && info->deriver && isValidPath(*info->deriver))
enqueue(*info->deriver);
}
{
auto state(state_.lock());
assert(state->pending);
if (!--state->pending) done.notify_one();
}
2018-03-27 20:16:01 +00:00
} catch (...) {
auto state(state_.lock());
2018-03-27 20:16:01 +00:00
if (!state->exc) state->exc = std::current_exception();
assert(state->pending);
if (!--state->pending) done.notify_one();
2018-03-27 20:16:01 +00:00
};
}});
};
for (auto & startPath : startPaths)
enqueue(startPath);
{
auto state(state_.lock());
while (state->pending) state.wait(done);
if (state->exc) std::rethrow_exception(state->exc);
}
}
void Store::computeFSClosure(const StorePath & startPath,
StorePathSet & paths_, bool flipDirection, bool includeOutputs, bool includeDerivers)
{
StorePathSet paths;
paths.insert(startPath);
computeFSClosure(paths, paths_, flipDirection, includeOutputs, includeDerivers);
}
std::optional<ContentAddress> getDerivationCA(const BasicDerivation & drv)
{
auto out = drv.outputs.find("out");
if (out != drv.outputs.end()) {
2020-08-05 14:44:39 +00:00
if (auto v = std::get_if<DerivationOutputCAFixed>(&out->second.output))
return v->hash;
}
return std::nullopt;
}
void Store::queryMissing(const std::vector<BuildableReq> & targets,
StorePathSet & willBuild_, StorePathSet & willSubstitute_, StorePathSet & unknown_,
2020-07-30 11:10:49 +00:00
uint64_t & downloadSize_, uint64_t & narSize_)
{
Activity act(*logger, lvlDebug, actUnknown, "querying info about missing paths");
downloadSize_ = narSize_ = 0;
ThreadPool pool;
download-from-binary-cache: parallelise fetching of NAR info files Getting substitute information using the binary cache substituter has non-trivial latency overhead. A package or NixOS system configuration can have hundreds of dependencies, and in the worst case (when the local info cache is empty) we have to do a separate HTTP request for each of these. If the ping time to the server is t, getting N info files will take tN seconds; e.g., with a ping time of 0.1s to nixos.org, sequentially downloading 1000 info files (a typical NixOS config) will take at least 100 seconds. To fix this problem, the binary cache substituter can now perform requests in parallel. This required changing the substituter interface to support a function querySubstitutablePathInfos() that queries multiple paths at the same time, and rewriting queryMissing() to take advantage of parallelism. (Due to local caching, parallelising queryMissing() is sufficient for most use cases, since it's almost always called before building a derivation and thus fills the local info cache.) For example, parallelism speeds up querying all 1056 paths in a particular NixOS system configuration from 116s to 2.6s. It works so well because the eccentricity of the top-level derivation in the dependency graph is only 9. So we only need 10 round-trips (when using an unlimited number of parallel connections) to get everything. Currently we do a maximum of 150 parallel connections to the server. Thus it's important that the binary cache server (e.g. nixos.org) has a high connection limit. Alternatively we could use HTTP pipelining, but WWW::Curl doesn't support it and libcurl has a hard-coded limit of 5 requests per pipeline.
2012-07-06 23:08:20 +00:00
struct State
{
std::unordered_set<std::string> done;
StorePathSet & unknown, & willSubstitute, & willBuild;
2020-07-30 11:10:49 +00:00
uint64_t & downloadSize;
uint64_t & narSize;
};
download-from-binary-cache: parallelise fetching of NAR info files Getting substitute information using the binary cache substituter has non-trivial latency overhead. A package or NixOS system configuration can have hundreds of dependencies, and in the worst case (when the local info cache is empty) we have to do a separate HTTP request for each of these. If the ping time to the server is t, getting N info files will take tN seconds; e.g., with a ping time of 0.1s to nixos.org, sequentially downloading 1000 info files (a typical NixOS config) will take at least 100 seconds. To fix this problem, the binary cache substituter can now perform requests in parallel. This required changing the substituter interface to support a function querySubstitutablePathInfos() that queries multiple paths at the same time, and rewriting queryMissing() to take advantage of parallelism. (Due to local caching, parallelising queryMissing() is sufficient for most use cases, since it's almost always called before building a derivation and thus fills the local info cache.) For example, parallelism speeds up querying all 1056 paths in a particular NixOS system configuration from 116s to 2.6s. It works so well because the eccentricity of the top-level derivation in the dependency graph is only 9. So we only need 10 round-trips (when using an unlimited number of parallel connections) to get everything. Currently we do a maximum of 150 parallel connections to the server. Thus it's important that the binary cache server (e.g. nixos.org) has a high connection limit. Alternatively we could use HTTP pipelining, but WWW::Curl doesn't support it and libcurl has a hard-coded limit of 5 requests per pipeline.
2012-07-06 23:08:20 +00:00
struct DrvState
{
size_t left;
bool done = false;
StorePathSet outPaths;
DrvState(size_t left) : left(left) { }
};
download-from-binary-cache: parallelise fetching of NAR info files Getting substitute information using the binary cache substituter has non-trivial latency overhead. A package or NixOS system configuration can have hundreds of dependencies, and in the worst case (when the local info cache is empty) we have to do a separate HTTP request for each of these. If the ping time to the server is t, getting N info files will take tN seconds; e.g., with a ping time of 0.1s to nixos.org, sequentially downloading 1000 info files (a typical NixOS config) will take at least 100 seconds. To fix this problem, the binary cache substituter can now perform requests in parallel. This required changing the substituter interface to support a function querySubstitutablePathInfos() that queries multiple paths at the same time, and rewriting queryMissing() to take advantage of parallelism. (Due to local caching, parallelising queryMissing() is sufficient for most use cases, since it's almost always called before building a derivation and thus fills the local info cache.) For example, parallelism speeds up querying all 1056 paths in a particular NixOS system configuration from 116s to 2.6s. It works so well because the eccentricity of the top-level derivation in the dependency graph is only 9. So we only need 10 round-trips (when using an unlimited number of parallel connections) to get everything. Currently we do a maximum of 150 parallel connections to the server. Thus it's important that the binary cache server (e.g. nixos.org) has a high connection limit. Alternatively we could use HTTP pipelining, but WWW::Curl doesn't support it and libcurl has a hard-coded limit of 5 requests per pipeline.
2012-07-06 23:08:20 +00:00
Sync<State> state_(State{{}, unknown_, willSubstitute_, willBuild_, downloadSize_, narSize_});
download-from-binary-cache: parallelise fetching of NAR info files Getting substitute information using the binary cache substituter has non-trivial latency overhead. A package or NixOS system configuration can have hundreds of dependencies, and in the worst case (when the local info cache is empty) we have to do a separate HTTP request for each of these. If the ping time to the server is t, getting N info files will take tN seconds; e.g., with a ping time of 0.1s to nixos.org, sequentially downloading 1000 info files (a typical NixOS config) will take at least 100 seconds. To fix this problem, the binary cache substituter can now perform requests in parallel. This required changing the substituter interface to support a function querySubstitutablePathInfos() that queries multiple paths at the same time, and rewriting queryMissing() to take advantage of parallelism. (Due to local caching, parallelising queryMissing() is sufficient for most use cases, since it's almost always called before building a derivation and thus fills the local info cache.) For example, parallelism speeds up querying all 1056 paths in a particular NixOS system configuration from 116s to 2.6s. It works so well because the eccentricity of the top-level derivation in the dependency graph is only 9. So we only need 10 round-trips (when using an unlimited number of parallel connections) to get everything. Currently we do a maximum of 150 parallel connections to the server. Thus it's important that the binary cache server (e.g. nixos.org) has a high connection limit. Alternatively we could use HTTP pipelining, but WWW::Curl doesn't support it and libcurl has a hard-coded limit of 5 requests per pipeline.
2012-07-06 23:08:20 +00:00
std::function<void(BuildableReq)> doPath;
download-from-binary-cache: parallelise fetching of NAR info files Getting substitute information using the binary cache substituter has non-trivial latency overhead. A package or NixOS system configuration can have hundreds of dependencies, and in the worst case (when the local info cache is empty) we have to do a separate HTTP request for each of these. If the ping time to the server is t, getting N info files will take tN seconds; e.g., with a ping time of 0.1s to nixos.org, sequentially downloading 1000 info files (a typical NixOS config) will take at least 100 seconds. To fix this problem, the binary cache substituter can now perform requests in parallel. This required changing the substituter interface to support a function querySubstitutablePathInfos() that queries multiple paths at the same time, and rewriting queryMissing() to take advantage of parallelism. (Due to local caching, parallelising queryMissing() is sufficient for most use cases, since it's almost always called before building a derivation and thus fills the local info cache.) For example, parallelism speeds up querying all 1056 paths in a particular NixOS system configuration from 116s to 2.6s. It works so well because the eccentricity of the top-level derivation in the dependency graph is only 9. So we only need 10 round-trips (when using an unlimited number of parallel connections) to get everything. Currently we do a maximum of 150 parallel connections to the server. Thus it's important that the binary cache server (e.g. nixos.org) has a high connection limit. Alternatively we could use HTTP pipelining, but WWW::Curl doesn't support it and libcurl has a hard-coded limit of 5 requests per pipeline.
2012-07-06 23:08:20 +00:00
auto mustBuildDrv = [&](const StorePath & drvPath, const Derivation & drv) {
{
auto state(state_.lock());
state->willBuild.insert(drvPath);
}
for (auto & i : drv.inputDrvs)
pool.enqueue(std::bind(doPath, BuildableReqFromDrv { i.first, i.second }));
};
download-from-binary-cache: parallelise fetching of NAR info files Getting substitute information using the binary cache substituter has non-trivial latency overhead. A package or NixOS system configuration can have hundreds of dependencies, and in the worst case (when the local info cache is empty) we have to do a separate HTTP request for each of these. If the ping time to the server is t, getting N info files will take tN seconds; e.g., with a ping time of 0.1s to nixos.org, sequentially downloading 1000 info files (a typical NixOS config) will take at least 100 seconds. To fix this problem, the binary cache substituter can now perform requests in parallel. This required changing the substituter interface to support a function querySubstitutablePathInfos() that queries multiple paths at the same time, and rewriting queryMissing() to take advantage of parallelism. (Due to local caching, parallelising queryMissing() is sufficient for most use cases, since it's almost always called before building a derivation and thus fills the local info cache.) For example, parallelism speeds up querying all 1056 paths in a particular NixOS system configuration from 116s to 2.6s. It works so well because the eccentricity of the top-level derivation in the dependency graph is only 9. So we only need 10 round-trips (when using an unlimited number of parallel connections) to get everything. Currently we do a maximum of 150 parallel connections to the server. Thus it's important that the binary cache server (e.g. nixos.org) has a high connection limit. Alternatively we could use HTTP pipelining, but WWW::Curl doesn't support it and libcurl has a hard-coded limit of 5 requests per pipeline.
2012-07-06 23:08:20 +00:00
auto checkOutput = [&](
const StorePath & drvPath, ref<Derivation> drv, const StorePath & outPath, ref<Sync<DrvState>> drvState_)
{
if (drvState_->lock()->done) return;
download-from-binary-cache: parallelise fetching of NAR info files Getting substitute information using the binary cache substituter has non-trivial latency overhead. A package or NixOS system configuration can have hundreds of dependencies, and in the worst case (when the local info cache is empty) we have to do a separate HTTP request for each of these. If the ping time to the server is t, getting N info files will take tN seconds; e.g., with a ping time of 0.1s to nixos.org, sequentially downloading 1000 info files (a typical NixOS config) will take at least 100 seconds. To fix this problem, the binary cache substituter can now perform requests in parallel. This required changing the substituter interface to support a function querySubstitutablePathInfos() that queries multiple paths at the same time, and rewriting queryMissing() to take advantage of parallelism. (Due to local caching, parallelising queryMissing() is sufficient for most use cases, since it's almost always called before building a derivation and thus fills the local info cache.) For example, parallelism speeds up querying all 1056 paths in a particular NixOS system configuration from 116s to 2.6s. It works so well because the eccentricity of the top-level derivation in the dependency graph is only 9. So we only need 10 round-trips (when using an unlimited number of parallel connections) to get everything. Currently we do a maximum of 150 parallel connections to the server. Thus it's important that the binary cache server (e.g. nixos.org) has a high connection limit. Alternatively we could use HTTP pipelining, but WWW::Curl doesn't support it and libcurl has a hard-coded limit of 5 requests per pipeline.
2012-07-06 23:08:20 +00:00
SubstitutablePathInfos infos;
querySubstitutablePathInfos({{outPath, getDerivationCA(*drv)}}, infos);
if (infos.empty()) {
drvState_->lock()->done = true;
mustBuildDrv(drvPath, *drv);
} else {
{
auto drvState(drvState_->lock());
if (drvState->done) return;
assert(drvState->left);
drvState->left--;
drvState->outPaths.insert(outPath);
if (!drvState->left) {
for (auto & path : drvState->outPaths)
pool.enqueue(std::bind(doPath, BuildableOpaque { path } ));
download-from-binary-cache: parallelise fetching of NAR info files Getting substitute information using the binary cache substituter has non-trivial latency overhead. A package or NixOS system configuration can have hundreds of dependencies, and in the worst case (when the local info cache is empty) we have to do a separate HTTP request for each of these. If the ping time to the server is t, getting N info files will take tN seconds; e.g., with a ping time of 0.1s to nixos.org, sequentially downloading 1000 info files (a typical NixOS config) will take at least 100 seconds. To fix this problem, the binary cache substituter can now perform requests in parallel. This required changing the substituter interface to support a function querySubstitutablePathInfos() that queries multiple paths at the same time, and rewriting queryMissing() to take advantage of parallelism. (Due to local caching, parallelising queryMissing() is sufficient for most use cases, since it's almost always called before building a derivation and thus fills the local info cache.) For example, parallelism speeds up querying all 1056 paths in a particular NixOS system configuration from 116s to 2.6s. It works so well because the eccentricity of the top-level derivation in the dependency graph is only 9. So we only need 10 round-trips (when using an unlimited number of parallel connections) to get everything. Currently we do a maximum of 150 parallel connections to the server. Thus it's important that the binary cache server (e.g. nixos.org) has a high connection limit. Alternatively we could use HTTP pipelining, but WWW::Curl doesn't support it and libcurl has a hard-coded limit of 5 requests per pipeline.
2012-07-06 23:08:20 +00:00
}
}
download-from-binary-cache: parallelise fetching of NAR info files Getting substitute information using the binary cache substituter has non-trivial latency overhead. A package or NixOS system configuration can have hundreds of dependencies, and in the worst case (when the local info cache is empty) we have to do a separate HTTP request for each of these. If the ping time to the server is t, getting N info files will take tN seconds; e.g., with a ping time of 0.1s to nixos.org, sequentially downloading 1000 info files (a typical NixOS config) will take at least 100 seconds. To fix this problem, the binary cache substituter can now perform requests in parallel. This required changing the substituter interface to support a function querySubstitutablePathInfos() that queries multiple paths at the same time, and rewriting queryMissing() to take advantage of parallelism. (Due to local caching, parallelising queryMissing() is sufficient for most use cases, since it's almost always called before building a derivation and thus fills the local info cache.) For example, parallelism speeds up querying all 1056 paths in a particular NixOS system configuration from 116s to 2.6s. It works so well because the eccentricity of the top-level derivation in the dependency graph is only 9. So we only need 10 round-trips (when using an unlimited number of parallel connections) to get everything. Currently we do a maximum of 150 parallel connections to the server. Thus it's important that the binary cache server (e.g. nixos.org) has a high connection limit. Alternatively we could use HTTP pipelining, but WWW::Curl doesn't support it and libcurl has a hard-coded limit of 5 requests per pipeline.
2012-07-06 23:08:20 +00:00
}
};
download-from-binary-cache: parallelise fetching of NAR info files Getting substitute information using the binary cache substituter has non-trivial latency overhead. A package or NixOS system configuration can have hundreds of dependencies, and in the worst case (when the local info cache is empty) we have to do a separate HTTP request for each of these. If the ping time to the server is t, getting N info files will take tN seconds; e.g., with a ping time of 0.1s to nixos.org, sequentially downloading 1000 info files (a typical NixOS config) will take at least 100 seconds. To fix this problem, the binary cache substituter can now perform requests in parallel. This required changing the substituter interface to support a function querySubstitutablePathInfos() that queries multiple paths at the same time, and rewriting queryMissing() to take advantage of parallelism. (Due to local caching, parallelising queryMissing() is sufficient for most use cases, since it's almost always called before building a derivation and thus fills the local info cache.) For example, parallelism speeds up querying all 1056 paths in a particular NixOS system configuration from 116s to 2.6s. It works so well because the eccentricity of the top-level derivation in the dependency graph is only 9. So we only need 10 round-trips (when using an unlimited number of parallel connections) to get everything. Currently we do a maximum of 150 parallel connections to the server. Thus it's important that the binary cache server (e.g. nixos.org) has a high connection limit. Alternatively we could use HTTP pipelining, but WWW::Curl doesn't support it and libcurl has a hard-coded limit of 5 requests per pipeline.
2012-07-06 23:08:20 +00:00
doPath = [&](const BuildableReq & req) {
{
auto state(state_.lock());
if (!state->done.insert(req.to_string(*this)).second) return;
}
download-from-binary-cache: parallelise fetching of NAR info files Getting substitute information using the binary cache substituter has non-trivial latency overhead. A package or NixOS system configuration can have hundreds of dependencies, and in the worst case (when the local info cache is empty) we have to do a separate HTTP request for each of these. If the ping time to the server is t, getting N info files will take tN seconds; e.g., with a ping time of 0.1s to nixos.org, sequentially downloading 1000 info files (a typical NixOS config) will take at least 100 seconds. To fix this problem, the binary cache substituter can now perform requests in parallel. This required changing the substituter interface to support a function querySubstitutablePathInfos() that queries multiple paths at the same time, and rewriting queryMissing() to take advantage of parallelism. (Due to local caching, parallelising queryMissing() is sufficient for most use cases, since it's almost always called before building a derivation and thus fills the local info cache.) For example, parallelism speeds up querying all 1056 paths in a particular NixOS system configuration from 116s to 2.6s. It works so well because the eccentricity of the top-level derivation in the dependency graph is only 9. So we only need 10 round-trips (when using an unlimited number of parallel connections) to get everything. Currently we do a maximum of 150 parallel connections to the server. Thus it's important that the binary cache server (e.g. nixos.org) has a high connection limit. Alternatively we could use HTTP pipelining, but WWW::Curl doesn't support it and libcurl has a hard-coded limit of 5 requests per pipeline.
2012-07-06 23:08:20 +00:00
std::visit(overloaded {
[&](BuildableReqFromDrv bfd) {
if (!isValidPath(bfd.drvPath)) {
// FIXME: we could try to substitute the derivation.
auto state(state_.lock());
state->unknown.insert(bfd.drvPath);
return;
}
StorePathSet invalid;
2020-08-07 19:09:26 +00:00
/* true for regular derivations, and CA derivations for which we
have a trust mapping for all wanted outputs. */
auto knownOutputPaths = true;
for (auto & [outputName, pathOpt] : queryPartialDerivationOutputMap(bfd.drvPath)) {
2020-08-07 19:09:26 +00:00
if (!pathOpt) {
knownOutputPaths = false;
break;
}
if (wantOutput(outputName, bfd.outputs) && !isValidPath(*pathOpt))
invalid.insert(*pathOpt);
2020-08-07 19:09:26 +00:00
}
if (knownOutputPaths && invalid.empty()) return;
auto drv = make_ref<Derivation>(derivationFromPath(bfd.drvPath));
ParsedDerivation parsedDrv(StorePath(bfd.drvPath), *drv);
2020-08-07 19:09:26 +00:00
if (knownOutputPaths && settings.useSubstitutes && parsedDrv.substitutesAllowed()) {
auto drvState = make_ref<Sync<DrvState>>(DrvState(invalid.size()));
for (auto & output : invalid)
pool.enqueue(std::bind(checkOutput, bfd.drvPath, drv, output, drvState));
download-from-binary-cache: parallelise fetching of NAR info files Getting substitute information using the binary cache substituter has non-trivial latency overhead. A package or NixOS system configuration can have hundreds of dependencies, and in the worst case (when the local info cache is empty) we have to do a separate HTTP request for each of these. If the ping time to the server is t, getting N info files will take tN seconds; e.g., with a ping time of 0.1s to nixos.org, sequentially downloading 1000 info files (a typical NixOS config) will take at least 100 seconds. To fix this problem, the binary cache substituter can now perform requests in parallel. This required changing the substituter interface to support a function querySubstitutablePathInfos() that queries multiple paths at the same time, and rewriting queryMissing() to take advantage of parallelism. (Due to local caching, parallelising queryMissing() is sufficient for most use cases, since it's almost always called before building a derivation and thus fills the local info cache.) For example, parallelism speeds up querying all 1056 paths in a particular NixOS system configuration from 116s to 2.6s. It works so well because the eccentricity of the top-level derivation in the dependency graph is only 9. So we only need 10 round-trips (when using an unlimited number of parallel connections) to get everything. Currently we do a maximum of 150 parallel connections to the server. Thus it's important that the binary cache server (e.g. nixos.org) has a high connection limit. Alternatively we could use HTTP pipelining, but WWW::Curl doesn't support it and libcurl has a hard-coded limit of 5 requests per pipeline.
2012-07-06 23:08:20 +00:00
} else
mustBuildDrv(bfd.drvPath, *drv);
},
[&](BuildableOpaque bo) {
if (isValidPath(bo.path)) return;
SubstitutablePathInfos infos;
querySubstitutablePathInfos({{bo.path, std::nullopt}}, infos);
if (infos.empty()) {
auto state(state_.lock());
state->unknown.insert(bo.path);
return;
}
auto info = infos.find(bo.path);
assert(info != infos.end());
{
auto state(state_.lock());
state->willSubstitute.insert(bo.path);
state->downloadSize += info->second.downloadSize;
state->narSize += info->second.narSize;
}
for (auto & ref : info->second.references)
pool.enqueue(std::bind(doPath, BuildableOpaque { ref }));
},
}, req.raw());
};
for (auto & path : targets)
pool.enqueue(std::bind(doPath, path));
pool.process();
}
StorePaths Store::topoSortPaths(const StorePathSet & paths)
2011-12-30 15:02:50 +00:00
{
2020-07-27 20:45:34 +00:00
return topoSort(paths,
{[&](const StorePath & path) {
StorePathSet references;
try {
references = queryPathInfo(path)->references;
} catch (InvalidPath &) {
}
return references;
}},
{[&](const StorePath & path, const StorePath & parent) {
return BuildError(
"cycle detected in the references of '%s' from '%s'",
printStorePath(path),
printStorePath(parent));
}});
2011-12-30 15:02:50 +00:00
}
}