wassname/ray
1
0
Fork 0
mirror of https://github.com/wassname/ray.git synced 2026-06-28 20:56:34 +08:00
Code Issues Packages Projects Releases Wiki Activity

Fix autoscaler format string for memory (#5542)

Browse Source 
* add format string

* fix cast
This commit is contained in:
Eric Liang
2019-08-26 23:25:11 -07:00
committed by GitHub
parent ff73b67154
commit d20696300e
3 changed files with 24 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/ray/autoscaler/autoscaler.py
+11 -3
View File
@@ -28,7 +28,7 @@ from ray.autoscaler.updater import NodeUpdaterThread
from ray.ray_constants import AUTOSCALER_MAX_NUM_FAILURES, \
AUTOSCALER_MAX_LAUNCH_BATCH, AUTOSCALER_MAX_CONCURRENT_LAUNCHES, \
AUTOSCALER_UPDATE_INTERVAL_S, AUTOSCALER_HEARTBEAT_TIMEOUT_S, \
AUTOSCALER_RESOURCE_REQUEST_CHANNEL
AUTOSCALER_RESOURCE_REQUEST_CHANNEL, MEMORY_RESOURCE_UNIT_BYTES
from six import string_types
from six.moves import queue
@@ -254,11 +254,19 @@ class LoadMetrics(object):
ip: (now - t)
for ip, t in most_delayed_heartbeats
}
def format_resource(key, value):
if key in ["object_store_memory", "memory"]:
return "{} GiB".format(
round(value * MEMORY_RESOURCE_UNIT_BYTES / 1e9, 2))
else:
return round(value, 2)
return {
"ResourceUsage": ", ".join([
"{}/{} {}".format(
round(resources_used[rid], 2),
round(resources_total[rid], 2), rid)
format_resource(rid, resources_used[rid]),
format_resource(rid, resources_total[rid]), rid)
for rid in sorted(resources_used)
]),
"NumNodesConnected": len(self.static_resources_by_ip),
python/ray/services.py
+1 -1
View File
@@ -1301,7 +1301,7 @@ def _start_plasma_store(plasma_store_memory,
"plasma_directory argument must be provided.")
if not isinstance(plasma_store_memory, int):
raise Exception("plasma_store_memory should be an integer.")
plasma_store_memory = int(plasma_store_memory)
command = [
PLASMA_STORE_EXECUTABLE, "-s", socket_name, "-m",
python/ray/tests/test_autoscaler.py
+12 -3
View File
@@ -174,10 +174,19 @@ class LoadMetricsTest(unittest.TestCase):
lm = LoadMetrics()
lm.update("1.1.1.1", {"CPU": 2}, {"CPU": 0})
lm.update("2.2.2.2", {"CPU": 2, "GPU": 16}, {"CPU": 2, "GPU": 2})
lm.update("3.3.3.3", {
"memory": 20,
"object_store_memory": 40
}, {
"memory": 0,
"object_store_memory": 20
})
debug = lm.info_string()
assert "ResourceUsage=2.0/4.0 CPU, 14.0/16.0 GPU" in debug
assert "NumNodesConnected=2" in debug
assert "NumNodesUsed=1.88" in debug
assert ("ResourceUsage=2.0/4.0 CPU, 14.0/16.0 GPU, "
"1.05 GiB/1.05 GiB memory, "
"1.05 GiB/2.1 GiB object_store_memory") in debug
assert "NumNodesConnected=3" in debug
assert "NumNodesUsed=2.88" in debug
class AutoscalingTest(unittest.TestCase):