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.

1 files changed, 79 insertions, 27 deletions

diff --git a/src/libstore/misc.cc b/src/libstore/misc.cc
index 0934999363..aa5f6ff727 100644
--- a/src/libstore/misc.cc
+++ b/src/libstore/misc.cc
@@ -55,45 +55,97 @@ void queryMissing(StoreAPI & store, const PathSet & targets,
     
     PathSet todo(targets.begin(), targets.end()), done;
 
+    bool useSubstitutes = queryBoolSetting("build-use-substitutes", true);
+
+    /* Getting substitute info has high latency when using the binary
+       cache substituter.  Thus it's essential to do substitute
+       queries in parallel as much as possible.  To accomplish this
+       we do the following:
+
+       - For all paths still to be processed (‘todo’), we add all
+         paths for which we need info to the set ‘query’.  For an
+         unbuilt derivation this is the output paths; otherwise, it's
+         the path itself.
+
+       - We get info about all paths in ‘query’ in parallel.
+
+       - We process the results and add new items to ‘todo’ if
+         necessary.  E.g. if a path is substitutable, then we need to
+         get info on its references.
+
+       - Repeat until ‘todo’ is empty.
+    */
+
     while (!todo.empty()) {
-        Path p = *(todo.begin());
-        todo.erase(p);
-        if (done.find(p) != done.end()) continue;
-        done.insert(p);
-
-        if (isDerivation(p)) {
-            if (!store.isValidPath(p)) {
-                unknown.insert(p);
-                continue;
+              
+        PathSet query, todoDrv, todoNonDrv;
+
+        foreach (PathSet::iterator, i, todo) {
+            if (done.find(*i) != done.end()) continue;
+            done.insert(*i);
+
+            if (isDerivation(*i)) {
+                if (!store.isValidPath(*i)) {
+                    // FIXME: we could try to substitute p.
+                    unknown.insert(*i);
+                    continue;
+                }
+                Derivation drv = derivationFromPath(store, *i);
+
+                PathSet invalid;
+                foreach (DerivationOutputs::iterator, j, drv.outputs)
+                    if (!store.isValidPath(j->second.path)) invalid.insert(j->second.path);
+                if (invalid.empty()) continue;
+                
+                todoDrv.insert(*i);
+                if (useSubstitutes) query.insert(invalid.begin(), invalid.end());
+            }
+
+            else {
+                if (store.isValidPath(*i)) continue;
+                query.insert(*i);
+                todoNonDrv.insert(*i);
             }
-            Derivation drv = derivationFromPath(store, p);
+        }
+
+        todo.clear();
+        
+        SubstitutablePathInfos infos;
+        store.querySubstitutablePathInfos(query, infos);
+
+        foreach (PathSet::iterator, i, todoDrv) {
+            // FIXME: cache this
+            Derivation drv = derivationFromPath(store, *i);
 
             bool mustBuild = false;
-            foreach (DerivationOutputs::iterator, i, drv.outputs)
-                if (!store.isValidPath(i->second.path) &&
-                    !(queryBoolSetting("build-use-substitutes", true) && store.hasSubstitutes(i->second.path)))
-                    mustBuild = true;
+            if (useSubstitutes) {
+                foreach (DerivationOutputs::iterator, j, drv.outputs)
+                    if (!store.isValidPath(j->second.path) &&
+                        infos.find(j->second.path) == infos.end())
+                        mustBuild = true;
+            } else
+                mustBuild = true;
 
             if (mustBuild) {
-                willBuild.insert(p);
+                willBuild.insert(*i);
                 todo.insert(drv.inputSrcs.begin(), drv.inputSrcs.end());
                 foreach (DerivationInputs::iterator, i, drv.inputDrvs)
                     todo.insert(i->first);
-            } else 
+            } else
                 foreach (DerivationOutputs::iterator, i, drv.outputs)
-                    todo.insert(i->second.path);
+                    todoNonDrv.insert(i->second.path);
         }
-
-        else {
-            if (store.isValidPath(p)) continue;
-            SubstitutablePathInfo info;
-            if (store.querySubstitutablePathInfo(p, info)) {
-                willSubstitute.insert(p);
-                downloadSize += info.downloadSize;
-                narSize += info.narSize;
-                todo.insert(info.references.begin(), info.references.end());
+        
+        foreach (PathSet::iterator, i, todoNonDrv) {
+            done.insert(*i);
+            SubstitutablePathInfos::iterator info = infos.find(*i);
+            if (info != infos.end()) {
+                willSubstitute.insert(*i);
+                downloadSize += info->second.downloadSize;
+                narSize += info->second.narSize;
+                todo.insert(info->second.references.begin(), info->second.references.end());
             } else
-                unknown.insert(p);
+                unknown.insert(*i);
         }
     }
 }