Merge 1b69645641 into 479b6b73a9

doc/manual/source/advanced-topics/distributed-builds.md

@@ -108,3 +108,17 @@ causes the list of machines in `/etc/nix/machines` to be included.
 (This is the default.)
 
 [Nix instance]: @docroot@/glossary.md#gloss-nix-instance
+
+## Resource Management
+
+Adding `resource-management` to the `experimental-features` setting in `nix.conf` enables a basic resource management scheme for system features. This is akin to what can be accomplished with job schedulers like Slurm, where a remote machine can have a limited quantity of a resource that can be temporarily "consumed" by a job. This can be used with memory-heavy builds, or derivations that require exclusive access to particular hardware resources.
+
+Resource management is supported in both the supported features and mandatory features of a remote machine configuration, by appending a colon `:` to a feature name followed by the quantity that this machine has. This is tracked on a per-store basis, so different users on a multi-user installation share the same pool of resources for their remote build machines. A derivation specifies that it consumes a resource with the same notation in the `requiredSystemFeatures` attribute.
+
+For example, this builder can provide exclusive access to two GPUs and 128G of memory for remote builds:
+
+    builders = ssh://gpu-node x86_64-linux - 32 1 gpu:2,mem:128
+
+A derivation that might use this machine may set its `requiredSystemFeatures` to `["gpu:1" "mem:4"]` to indicate that it requires a GPU and consumes 4G of system memory. A particularly memory-heavy derivation that doesn't need a GPU may still use the machine with a value of `["mem:64"]`. This helps ensure that limited system resources are not over-consumed by remote builds. Note that Nix does not do any actual delegation or enforcement of GPU, memory, or other resource usage, that is up to the derivations to manage.
+
+When configuring the `system-features` setting on the remote machine's `nix.conf`, only include the name of the consumable feature, not the quantity availble. Resource limits are tracked on the dispatching end within the local store.

src/libstore/derivation-options.cc

@@ -6,6 +6,7 @@
 #include "nix/util/types.hh"
 #include "nix/util/util.hh"
 #include "nix/store/globals.hh"
+#include "nix/store/machines.hh"
 
 #include <optional>
 #include <string>
@@ -327,7 +328,14 @@ bool DerivationOptions::canBuildLocally(Store & localStore, const BasicDerivatio
     if (settings.maxBuildJobs.get() == 0 && !drv.isBuiltin())
         return false;
 
-    for (auto & feature : getRequiredSystemFeatures(drv))
+    auto features = getRequiredSystemFeatures(drv);
+    if (experimentalFeatureSettings.isEnabled(Xp::ResourceManagement)) {
+        auto featureCount = Machine::countFeatures(features);
+        for (auto & feature : featureCount)
+            if (!localStore.config.systemFeatures.get().count(feature.first))
+                return false;
+    } else
+        for (auto & feature : features)
             if (!localStore.config.systemFeatures.get().count(feature))
                 return false;
 

src/libstore/include/nix/store/machines.hh

@@ -12,6 +12,8 @@ struct Machine;
 
 typedef std::vector<Machine> Machines;
 
+typedef std::map<std::string, unsigned long> FeatureCount;
+
 struct Machine
 {
 
@@ -21,7 +23,9 @@ struct Machine
     const unsigned int maxJobs;
     const float speedFactor;
     const StringSet supportedFeatures;
+    const FeatureCount supportedFeaturesCount;
     const StringSet mandatoryFeatures;
+    const FeatureCount mandatoryFeaturesCount;
     const std::string sshPublicHostKey;
     bool enabled = true;
 
@@ -77,6 +81,11 @@ struct Machine
      * the same format.
      */
     static Machines parseConfig(const StringSet & defaultSystems, const std::string & config);
+
+    /**
+     * Count the number of each feature specified in a feature string.
+     */
+    static FeatureCount countFeatures(const StringSet & features);
 };
 
 /**

src/libstore/machines.cc

@@ -31,7 +31,9 @@ Machine::Machine(
     , maxJobs(maxJobs)
     , speedFactor(speedFactor == 0.0f ? 1.0f : speedFactor)
     , supportedFeatures(supportedFeatures)
+    , supportedFeaturesCount(countFeatures(supportedFeatures))
     , mandatoryFeatures(mandatoryFeatures)
+    , mandatoryFeaturesCount(countFeatures(mandatoryFeatures))
     , sshPublicHostKey(sshPublicHostKey)
 {
     if (speedFactor < 0.0)
@@ -45,16 +47,48 @@ bool Machine::systemSupported(const std::string & system) const
 
 bool Machine::allSupported(const StringSet & features) const
 {
+    if (experimentalFeatureSettings.isEnabled(Xp::ResourceManagement)) {
+        auto featuresCount = countFeatures(features);
+        return std::all_of(featuresCount.begin(), featuresCount.end(), [&](const auto & f) {
+            return (
+                supportedFeaturesCount.count(f.first) > 0 && (         // feature is supported, and
+                    supportedFeaturesCount.at(f.first) >= f.second ||  // we have the quantity of it needed or
+                    supportedFeaturesCount.at(f.first) == 0            // we have a limitless supply of it
+                )
+            ) || (
+                mandatoryFeaturesCount.count(f.first) > 0 && (
+                    mandatoryFeaturesCount.at(f.first) >= f.second ||
+                    mandatoryFeaturesCount.at(f.first) == 0
+                )
+            );
+        });
+    } else {
         return std::all_of(features.begin(), features.end(), [&](const std::string & feature) {
             return supportedFeatures.count(feature) || mandatoryFeatures.count(feature);
         });
+    }
 }
 
 bool Machine::mandatoryMet(const StringSet & features) const
 {
+    if (experimentalFeatureSettings.isEnabled(Xp::ResourceManagement)) {
+        auto featureCount = countFeatures(features);
+        return std::all_of(mandatoryFeaturesCount.begin(), mandatoryFeaturesCount.end(), [&](const auto & feature) {
+            return featureCount.count(feature.first);
+        });
+    } else {
         return std::all_of(mandatoryFeatures.begin(), mandatoryFeatures.end(), [&](const std::string & feature) {
             return features.count(feature);
         });
+    }
+}
+
+std::string escapeUri(std::string uri)
+{
+    if (uri.find(':') != std::string::npos) {
+        uri.replace(uri.find(':'), 3, "%3A");
+    }
+    return uri;
 }
 
 StoreReference Machine::completeStoreReference() const
@@ -81,7 +115,7 @@ StoreReference Machine::completeStoreReference() const
             for (auto & f : feats) {
                 if (fs.size() > 0)
                     fs += ' ';
-                fs += f;
+                fs += escapeUri(f);
             }
         };
         append(supportedFeatures);
@@ -207,6 +241,26 @@ Machines Machine::parseConfig(const StringSet & defaultSystems, const std::strin
     return parseBuilderLines(defaultSystems, builderLines);
 }
 
+FeatureCount Machine::countFeatures(const StringSet & features)
+{
+    FeatureCount fc;
+    for (auto & f : features) {
+        std::istringstream fss(f);
+        std::string name;
+        std::string quantity;
+        unsigned long ulquantity = 0;
+        std::getline(fss, name, ':');
+        if (std::getline(fss, quantity)) {
+            ulquantity = std::stoul(quantity);
+            if (ulquantity == 0) {
+                throw UsageError("quantity for feature %s must be > 0", name);
+            }
+        }
+        fc.emplace(name, ulquantity);
+    };
+    return fc;
+}
+
 Machines getMachines()
 {
     return Machine::parseConfig({settings.thisSystem}, settings.builders);

src/libutil/experimental-features.cc

@@ -25,7 +25,7 @@ struct ExperimentalFeatureDetails
  * feature, we either have no issue at all if few features are not added
  * at the end of the list, or a proper merge conflict if they are.
  */
-constexpr size_t numXpFeatures = 1 + static_cast<size_t>(Xp::BLAKE3Hashes);
+constexpr size_t numXpFeatures = 1 + static_cast<size_t>(Xp::ResourceManagement);
 
 constexpr std::array<ExperimentalFeatureDetails, numXpFeatures> xpFeatureDetails = {{
     {
@@ -321,6 +321,14 @@ constexpr std::array<ExperimentalFeatureDetails, numXpFeatures> xpFeatureDetails
         )",
         .trackingUrl = "",
     },
+    {
+        .tag = Xp::ResourceManagement,
+        .name = "resource-management",
+        .description = R"(
+            Enables support for resource management in remote build system features.
+        )",
+        .trackingUrl = "",
+    }
 }};
 
 static_assert(

src/libutil/include/nix/util/experimental-features.hh

@@ -39,6 +39,7 @@ enum struct ExperimentalFeature {
     PipeOperators,
     ExternalBuilders,
     BLAKE3Hashes,
+    ResourceManagement,
 };
 
 /**

src/nix/build-remote/build-remote.cc

@@ -37,9 +37,14 @@ std::string escapeUri(std::string uri)
 
 static std::string currentLoad;
 
-static AutoCloseFD openSlotLock(const Machine & m, uint64_t slot)
+static AutoCloseFD openSlotLock(const std::string storeUri, uint64_t slot)
 {
-    return openLockFile(fmt("%s/%s-%d", currentLoad, escapeUri(m.storeUri.render()), slot), true);
+    return openLockFile(fmt("%s/%s-%d", currentLoad, escapeUri(storeUri), slot), true);
+}
+
+static AutoCloseFD openFeatureSlotLock(const std::string storeUri, const std::string feature, unsigned int slot)
+{
+    return openLockFile(fmt("%s/%s-%s-%d", currentLoad, escapeUri(storeUri), feature, slot), true);
 }
 
 static bool allSupportedLocally(Store & store, const StringSet & requiredFeatures)
@@ -50,6 +55,47 @@ static bool allSupportedLocally(Store & store, const StringSet & requiredFeature
     return true;
 }
 
+using FeatureSlotLocks = std::map<std::string, std::vector<AutoCloseFD>>;
+
+static bool tryReserveFeatures(
+    const Machine & m,
+    const FeatureCount requiredFeatures,
+    FeatureSlotLocks & featureSlotLocks
+) {
+    bool allSatisfied = true;
+    for (auto & f : requiredFeatures) {
+        if (!f.second) {
+            continue;
+        }
+        std::vector<AutoCloseFD> locks(f.second);
+        unsigned int numLocked = 0;
+        for (unsigned int s = 0;
+            numLocked < f.second && (
+                (m.supportedFeaturesCount.find(f.first) != m.supportedFeaturesCount.end() &&
+                s < m.supportedFeaturesCount.at(f.first)) ||
+                (m.mandatoryFeaturesCount.find(f.first) != m.mandatoryFeaturesCount.end() &&
+                s < m.mandatoryFeaturesCount.at(f.first))
+            ); ++s) {
+            auto lock = openFeatureSlotLock(m.storeUri.render(), f.first, s);
+            if (lockFile(lock.get(), ltWrite, false)) {
+                locks[numLocked] = std::move(lock);
+                ++numLocked;
+            }
+        }
+        if (numLocked < f.second) {
+            allSatisfied = false;
+            break;
+        }
+        auto & fslDest = featureSlotLocks[f.first];
+        fslDest.insert(fslDest.end(), std::make_move_iterator(locks.begin()),
+            std::make_move_iterator(locks.end()));
+    }
+    if (!allSatisfied) {
+        featureSlotLocks.clear();
+    }
+    return allSatisfied;
+}
+
 static int main_build_remote(int argc, char ** argv)
 {
     {
@@ -93,6 +139,7 @@ static int main_build_remote(int argc, char ** argv)
 
         std::shared_ptr<Store> sshStore;
         AutoCloseFD bestSlotLock;
+        FeatureSlotLocks bestFeatureSlotLocks;
 
         auto machines = getMachines();
         debug("got %d remote builders", machines.size());
@@ -119,6 +166,12 @@ static int main_build_remote(int argc, char ** argv)
             auto neededSystem = readString(source);
             drvPath = store->parseStorePath(readString(source));
             auto requiredFeatures = readStrings<StringSet>(source);
+            auto requiredFeaturesCount = Machine::countFeatures(requiredFeatures);
+            bool needsResourceManagement = 0 < std::accumulate(
+                requiredFeaturesCount.begin(), requiredFeaturesCount.end(), 0,
+                [](auto total, auto feature) {
+                    return std::move(total) + feature.second;
+                });
 
             /* It would be possible to build locally after some builds clear out,
                so don't show the warning now: */
@@ -150,7 +203,7 @@ static int main_build_remote(int argc, char ** argv)
                         AutoCloseFD free;
                         uint64_t load = 0;
                         for (uint64_t slot = 0; slot < m.maxJobs; ++slot) {
-                            auto slotLock = openSlotLock(m, slot);
+                            auto slotLock = openSlotLock(m.storeUri.render(), slot);
                             if (lockFile(slotLock.get(), ltWrite, false)) {
                                 if (!free) {
                                     free = std::move(slotLock);
@@ -162,6 +215,13 @@ static int main_build_remote(int argc, char ** argv)
                         if (!free) {
                             continue;
                         }
+                        FeatureSlotLocks featureSlotLocks;
+                        if (needsResourceManagement &&
+                            experimentalFeatureSettings.isEnabled(Xp::ResourceManagement)) {
+                            if (!tryReserveFeatures(m, requiredFeaturesCount, featureSlotLocks)) {
+                                continue;
+                            }
+                        }
                         bool best = false;
                         if (!bestSlotLock) {
                             best = true;
@@ -179,6 +239,7 @@ static int main_build_remote(int argc, char ** argv)
                         if (best) {
                             bestLoad = load;
                             bestSlotLock = std::move(free);
+                            bestFeatureSlotLocks = std::move(featureSlotLocks);
                             bestMachine = &m;
                         }
                     }

tests/functional/build-remote-resource-management-should-fail.sh

@@ -0,0 +1,18 @@
+#!/usr/bin/env bash
+
+source common.sh
+
+enableFeatures "resource-management"
+
+requireSandboxSupport
+[[ $busybox =~ busybox ]] || skipTest "no busybox"
+
+here=$(readlink -f "$(dirname "${BASH_SOURCE[0]}")")
+export NIX_USER_CONF_FILES=$here/config/nix-with-resource-management.conf
+
+expectStderr 1 nix build -Lvf resource-management.nix \
+  --arg busybox "$busybox" \
+  --out-link "$TEST_ROOT/result-from-remote" \
+  --store "$TEST_ROOT/local" \
+  --builders "ssh-ng://localhost?system-features=testf - - 4 1 testf:1" \
+| grepQuiet "Failed to find a machine for remote build!"

tests/functional/build-remote-resource-management.sh

@@ -0,0 +1,19 @@
+#!/usr/bin/env bash
+
+source common.sh
+
+enableFeatures "resource-management"
+
+requireSandboxSupport
+[[ $busybox =~ busybox ]] || skipTest "no busybox"
+
+here=$(readlink -f "$(dirname "${BASH_SOURCE[0]}")")
+export NIX_USER_CONF_FILES=$here/config/nix-with-resource-management.conf
+
+nix build -Lvf resource-management.nix \
+  --arg busybox "$busybox" \
+  --out-link "$TEST_ROOT/result-from-remote" \
+  --store "$TEST_ROOT/local" \
+  --builders "ssh-ng://localhost?system-features=test - - 4 1 test:4"
+
+grepQuiet 'Hello World!' < "$TEST_ROOT/result-from-remote/hello"

tests/functional/config/nix-with-resource-management.conf

@@ -0,0 +1,2 @@
+experimental-features = resource-management nix-command
+system-features = test

tests/functional/meson.build

@@ -109,6 +109,8 @@ suites = [
       'build-remote-trustless-should-pass-3.sh',
       'build-remote-trustless-should-fail-0.sh',
       'build-remote-with-mounted-ssh-ng.sh',
+      'build-remote-resource-management-should-fail.sh',
+      'build-remote-resource-management.sh',
       'nar-access.sh',
       'impure-eval.sh',
       'pure-eval.sh',

tests/functional/resource-management.nix

@@ -0,0 +1,50 @@
+{ busybox }:
+
+with import ./config.nix;
+
+let
+  drv1 = mkDerivation {
+    name = "resource-management-1";
+    shell = busybox;
+    builder = ./simple.builder.sh;
+    PATH = "";
+    goodPath = path;
+    requiredSystemFeatures = ["test:2"];
+    meta.position = "${__curPos.file}:${toString __curPos.line}";
+  };
+  drv2 = mkDerivation {
+    name = "resource-management-2";
+    shell = busybox;
+    builder = ./simple.builder.sh;
+    PATH = "";
+    goodPath = path;
+    requiredSystemFeatures = ["test:2"];
+    meta.position = "${__curPos.file}:${toString __curPos.line}";
+  };
+  drv3 = mkDerivation {
+    name = "resource-management-3";
+    shell = busybox;
+    builder = ./simple.builder.sh;
+    PATH = "";
+    goodPath = path;
+    requiredSystemFeatures = ["test:2"];
+    meta.position = "${__curPos.file}:${toString __curPos.line}";
+  };
+  drv4 = mkDerivation {
+    name = "resource-management-4";
+    shell = busybox;
+    builder = ./simple.builder.sh;
+    PATH = "";
+    goodPath = path;
+    requiredSystemFeatures = ["test:2"];
+    meta.position = "${__curPos.file}:${toString __curPos.line}";
+  };
+in mkDerivation {
+  name = "resource-management";
+  shell = busybox;
+  builder = ./simple.builder.sh;
+  PATH = "";
+  goodPath = path;
+  DRVS = "${drv1}${drv2}${drv3}${drv4}";
+  meta.position = "${__curPos.file}:${toString __curPos.line}";
+}