From fda2c828987ce6855d0ea248f9abba3f168d45e9 Mon Sep 17 00:00:00 2001
From: Mirek Kratochvil <exa.exa@gmail.com>
Date: Fri, 18 Mar 2022 17:18:42 +0100
Subject: [PATCH] avoid using `@sync_add` on remotecalls

It seems like @sync_add adds the Futures to a queue (Channel) for @sync, which
in turn calls wait() for all the futures synchronously. Not only that is
slightly detrimental for network operations (latencies add up), but in case of
Distributed the call to wait() may actually cause some compilation on remote
processes, which is also wait()ed for. In result, some operations took a great
amount of "serial" processing time if executed on many workers at once.

For me, this closes #44645.

The major change can be illustrated as follows: First add some workers:

```
using Distributed
addprocs(10)
```

and then trigger something that, for example, causes package imports on the
workers:

```
using SomeTinyPackage
```

In my case (importing UnicodePlots on 10 workers), this improves the loading
time over 10 workers from ~11s to ~5.5s.

This is a far bigger issue when worker count gets high. The time of the
processing on each worker is usually around 0.3s, so triggering this problem
even on a relatively small cluster (64 workers) causes a really annoying delay,
and running `@everywhere` for the first time on reasonable clusters (I tested
with 1024 workers, see #44645) usually takes more than 5 minutes. Which sucks.

Anyway, on 64 workers this reduces the "first import" time from ~30s to ~6s,
and on 1024 workers this seems to reduce the time from over 5 minutes (I didn't
bother to measure that precisely now, sorry) to ~11s.

Related issues:
- Probably fixes #39291.
- #42156 is a kinda complementary -- it removes the most painful source of
  slowness (the 0.3s precompilation on the workers), but the fact that the
  wait()ing is serial remains a problem if the network latencies are high.
---
 stdlib/Distributed/src/Distributed.jl      | 4 ++--
 stdlib/Distributed/src/clusterserialize.jl | 2 +-
 stdlib/Distributed/src/macros.jl           | 4 ++--
 3 files changed, 5 insertions(+), 5 deletions(-)

diff --git a/stdlib/Distributed/src/Distributed.jl b/stdlib/Distributed/src/Distributed.jl
index d428a6df0e683..4a4b27350589d 100644
--- a/stdlib/Distributed/src/Distributed.jl
+++ b/stdlib/Distributed/src/Distributed.jl
@@ -10,7 +10,7 @@ import Base: getindex, wait, put!, take!, fetch, isready, push!, length,
              hash, ==, kill, close, isopen, showerror
 
 # imports for use
-using Base: Process, Semaphore, JLOptions, buffer_writes, @sync_add,
+using Base: Process, Semaphore, JLOptions, buffer_writes,
             VERSION_STRING, binding_module, atexit, julia_exename,
             julia_cmd, AsyncGenerator, acquire, release, invokelatest,
             shell_escape_posixly, shell_escape_csh,
@@ -76,7 +76,7 @@ function _require_callback(mod::Base.PkgId)
         # broadcast top-level (e.g. from Main) import/using from node 1 (only)
         @sync for p in procs()
             p == 1 && continue
-            @sync_add remotecall(p) do
+            @async remotecall_wait(p) do
                 Base.require(mod)
                 nothing
             end
diff --git a/stdlib/Distributed/src/clusterserialize.jl b/stdlib/Distributed/src/clusterserialize.jl
index e37987c5bf875..ce71081ba568e 100644
--- a/stdlib/Distributed/src/clusterserialize.jl
+++ b/stdlib/Distributed/src/clusterserialize.jl
@@ -243,7 +243,7 @@ An exception is raised if a global constant is requested to be cleared.
 """
 function clear!(syms, pids=workers(); mod=Main)
     @sync for p in pids
-        @sync_add remotecall(clear_impl!, p, syms, mod)
+        @async remotecall_wait(clear_impl!, p, syms, mod)
     end
 end
 clear!(sym::Symbol, pid::Int; mod=Main) = clear!([sym], [pid]; mod=mod)
diff --git a/stdlib/Distributed/src/macros.jl b/stdlib/Distributed/src/macros.jl
index 0a62fdd5439f0..5dd622827b6f9 100644
--- a/stdlib/Distributed/src/macros.jl
+++ b/stdlib/Distributed/src/macros.jl
@@ -222,10 +222,10 @@ function remotecall_eval(m::Module, procs, ex)
             if pid == myid()
                 run_locally += 1
             else
-                @sync_add remotecall(Core.eval, pid, m, ex)
+                @async remotecall_wait(Core.eval, pid, m, ex)
             end
         end
-        yield() # ensure that the remotecall_fetch have had a chance to start
+        yield() # ensure that the remotecalls have had a chance to start
 
         # execute locally last as we do not want local execution to block serialization
         # of the request to remote nodes.