spawn many isolates: better errors if too many, scale up

[sigmundch]

I'm creating this bug to track progress on this issue:
 - we need to report better error messages when we can't create another isolate (e.g. an exception like 'TooManyIsolates')
 - we eventually want to scale to a larger number of isolates.

---

(original bug as reported in misc@)

I can only spawn 79 isolate, is there any bug in below code or any
limitation on dart vm ?

What steps will reproduce the problem?

run below script on Linux with latest dart SDK.

import('dart:isolate');

myfun() {
 port.receive( (msg, SendPort reply) {
   print("recieve: $msg");
   reply.send("received");
 });
}

void test() {

 SendPort port = spawnFunction( myfun );
 port.call("hello").then( (reply) {
   print(reply);
 });

}

main()
{

 var times = 80;

 Stopwatch w = new Stopwatch();
 w.start();

 for(int i=0; i<times; i++) {
   test();
 }
 w.stop();

 print("time used: ${w.elapsedInMs()}");

}

./runtime/vm/scavenger.cc:162: error: expected: space_ != NULL

155 Scavenger::Scavenger(Heap* heap, intptr_t max_capacity, uword
object_alignment)
156 : heap_(heap),
157 object_alignment_(object_alignment),
158 count_(0),
159 scavenging_(false) {
160 // Allocate the virtual memory for this scavenge heap.
161 space_ = VirtualMemory::Reserve(max_capacity);
162 ASSERT(space_ != NULL);

ulimit -a
core file size (blocks, -c) 0
data seg size (kbytes, -d) unlimited
scheduling priority (-e) 20
file size (blocks, -f) unlimited
pending signals (-i) 16382
max locked memory (kbytes, -l) 64
max memory size (kbytes, -m) unlimited
open files (-n) 1024
pipe size (512 bytes, -p) 8
POSIX message queues (bytes, -q) 819200
real-time priority (-r) 0
stack size (kbytes, -s) 8192
cpu time (seconds, -t) unlimited
max user processes (-u) unlimited
virtual memory (kbytes, -v) unlimited
file locks (-x) unlimited

[kasperl]

I'm marking this as a VM issue for now.

---

Added Area-VM, Triaged labels.

[sigmundch]

Added Isolates label.

[iposva-google]

Set owner to @mraleph.

[sigmundch]

Added Library-Isolates label.

[sigmundch]

Removed Isolates label.

[mraleph]

It is still reproducible, though now I can have ~130 isolates.

I am not sure if getting an error instead of a crash should be scheduled for M1.

What do you think Todd?

---

Set owner to @turnidge.

[dgrove]

Added Library-Isolate label.

[dgrove]

Removed Library-Isolates label.

[iposva-google]

Removed the owner.
Added this to the Later milestone.
Removed Library-Isolate label.

[iposva-google]

Removed Priority-Medium label.
Added Priority-Unassigned label.

[iposva-google]

Waiting for the new isolate library, which should allow error reporting when spawning isolates.

[DartBot]

This comment was originally written by [email protected]

---

I found this benchmark of isolate performance in the misc group https://groups.google.com/a/dartlang.org/forum/#!msg/vm-dev/WxhM23rYhig/WNiGHYkBudcJ.

I performed a basic benchmark to explore the performance of an isolate per request model. In dart it takes between 22 and 30 ms from the time Isolate.spawn is called to the time the entryPoint method is actually called (measured by sending new DateTime.now to the child isolate). The equivalent java version takes 1 ms or less. The basic gist is that the creating isolates is likely too costly for a per request model (where the servers portion of response budget is < 100 ms). I did a basic warm up for both vm's spinning up 1000 isolates/threads.

[kasperl]

Removed this from the Later milestone.
Added Oldschool-Milestone-Later label.

[kasperl]

Removed Oldschool-Milestone-Later label.