Skip to content

Commit 86e1326

Browse files
Frnd-meFrnd-me
authored
fix: fixing deployment file (#14)
1 parent 1d9c3a6 commit 86e1326

5 files changed

Lines changed: 1578 additions & 137 deletions

File tree

‎config/redis/components-grid5000/.gitignore‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1 @@
1+
pubsub.yaml

‎config/redis/components-grid5000/pubsub.yaml‎

Lines changed: 0 additions & 12 deletions
This file was deleted.

‎diningPhilosophers/deploy.py‎

Lines changed: 128 additions & 125 deletions
Original file line numberDiff line numberDiff line change
@@ -1,29 +1,29 @@
1+
import enoslib as en
12
import logging
2-
import time
33
import os
44
import tarfile
5+
import time
56
from pathlib import Path
6-
import enoslib as en
77
from tqdm import tqdm
88

99
en.init_logging(level=logging.INFO)
1010

1111
# --- Experiment parameters ---
12-
ACTOR_IMAGE = "collaborativestatemachines/cirrina-baselines-diningPhilosophers"
12+
ACTOR_IMAGE = "collaborativestatemachines/cirrina-baselines-diningphilosophers:unstable"
1313
REDIS_IMAGE = "redis:8.2.4-alpine"
1414
SIDECAR_IMAGE = "daprio/daprd:edge"
1515
PLACEMENT_IMAGE = "daprio/placement:1.16.0"
1616
COMPONENTS_PATH = "/tmp/dapr-components"
17-
LOCAL_ROOT = Path("./results")
17+
LOCAL_ROOT = Path("./results")
1818
TIME_BEFORE_FETCH = 60 * 20
19-
NUM_RUNS = 5
19+
NUM_RUNS = 5
2020
# -------------------------------------
2121

2222
# Infrastructure Reservation
2323
conf = (
24-
en.G5kConf.from_settings(job_name=Path(__file__).name, walltime="03:00:00")
25-
.add_machine(roles=["arbitrator"], cluster="gros", nodes=1)
26-
.add_machine(roles=["worker"], cluster="gros", nodes=6)
24+
en.G5kConf.from_settings(job_name=Path(__file__).name, walltime="03:00:00")
25+
.add_machine(roles=["arbitrator"], cluster="gros", nodes=1)
26+
.add_machine(roles=["worker"], cluster="gros", nodes=6)
2727
)
2828

2929
provider = en.G5k(conf)
@@ -56,7 +56,7 @@
5656
value: ""
5757
"""
5858

59-
Path("../config/redis/components-grid5000/pubsub.yaml").write_text(pubsub_yaml)
59+
Path("config/redis/components-grid5000/pubsub.yaml").write_text(pubsub_yaml)
6060

6161
# Network emulation
6262
netem = en.NetemHTB()
@@ -67,121 +67,124 @@
6767
netem.deploy()
6868

6969
for run_idx in range(1, NUM_RUNS + 1):
70-
run_label = f"run{run_idx}"
71-
print(f"\n>>> Starting {run_label}...")
72-
73-
# Ensure a fresh metrics directory on every node and push component files
74-
with en.actions(roles=roles) as a:
75-
a.file(path="/tmp/metrics", state="absent")
76-
a.file(path="/tmp/metrics", state="directory", mode="0777")
77-
a.file(path=COMPONENTS_PATH, state="directory")
78-
a.copy(src="../config/redis/components-grid5000/", dest=COMPONENTS_PATH + "/")
79-
80-
# Deploy Containers on Arbitrator
81-
with en.actions(roles=roles["arbitrator"]) as a:
82-
a.docker_container(
83-
name="redis", image=REDIS_IMAGE,
84-
network_mode="host", state="started"
85-
)
86-
a.docker_container(
87-
name="placement", image=PLACEMENT_IMAGE,
88-
network_mode="host", state="started",
89-
command=["./placement", "--port", "50006"]
90-
)
91-
a.docker_container(
92-
name="arbitrator-sidecar", image=SIDECAR_IMAGE,
93-
network_mode="host", state="started",
94-
command=[
95-
"./daprd",
96-
"--app-id", "arbitrator-sidecar",
97-
"--app-port", "3000",
98-
"--resources-path", "/components",
99-
"--placement-host-address", "localhost:50006",
100-
"--metrics-port", "9091",
101-
],
102-
volumes=[f"{COMPONENTS_PATH}:/components"],
103-
)
104-
a.docker_container(
105-
name="arbitrator", image=ACTOR_IMAGE,
106-
network_mode="host", state="started",
107-
volumes=["/tmp/metrics:/metrics:rw"],
108-
env={
109-
"ROLE": "arbitrator",
110-
"NUMBER_OF_PHILOSOPHERS": "6",
111-
"DAPR_HTTP_ENDPOINT": "http://localhost:3500",
112-
"DAPR_GRPC_ENDPOINT": "http://localhost:50001",
113-
"METRICS_DIRECTORY": "/metrics",
114-
},
115-
)
116-
117-
# Deploy Containers on Workers
118-
for i, host in enumerate(roles["worker"]):
119-
with en.actions(pattern_hosts=host.address, roles=roles) as a:
120-
a.docker_container(
121-
name="redis", image=REDIS_IMAGE,
122-
network_mode="host", state="started"
123-
)
124-
a.docker_container(
125-
name="placement", image=PLACEMENT_IMAGE,
126-
network_mode="host", state="started",
127-
command=["./placement", "--port", "50006"]
128-
)
129-
a.docker_container(
130-
name=f"w{i}-sidecar", image=SIDECAR_IMAGE,
131-
network_mode="host", state="started",
132-
command=[
133-
"./daprd",
134-
"--app-id", f"w{i}-sidecar",
135-
"--app-port", "3000",
136-
"--resources-path", "/components",
137-
"--placement-host-address", "localhost:50006",
138-
"--metrics-port", "9091",
139-
],
140-
volumes=[f"{COMPONENTS_PATH}:/components"],
141-
)
142-
a.docker_container(
143-
name=f"w{i}", image=ACTOR_IMAGE,
144-
network_mode="host", state="started",
145-
volumes=["/tmp/metrics:/metrics:rw"],
146-
env={
147-
"PHILOSOPHER_ID": str(i),
148-
"DAPR_HTTP_ENDPOINT": "http://localhost:3500",
149-
"DAPR_GRPC_ENDPOINT": "http://localhost:50001",
150-
"METRICS_DIRECTORY": "/metrics",
151-
},
152-
)
153-
154-
# Wait for data collection
155-
print(f"--- {run_label}: Collecting data for {TIME_BEFORE_FETCH}s ---")
156-
for _ in tqdm(range(TIME_BEFORE_FETCH), desc=run_label, unit="s", mininterval=60):
157-
time.sleep(1)
158-
159-
# Fetch and Organize Results locally into run1, run2, etc.
160-
run_dest = LOCAL_ROOT / run_label
161-
run_dest.mkdir(parents=True, exist_ok=True)
162-
163-
with en.actions(roles=roles) as a:
164-
a.archive(path="/tmp/metrics", dest="/tmp/metrics.tar.gz", format="gz")
165-
a.fetch(src="/tmp/metrics.tar.gz", dest=str(run_dest), flat=False)
166-
167-
# Local extraction
168-
print(f"--- {run_label}: Extracting results ---")
169-
for host in en.get_hosts(roles):
170-
host_dir = run_dest / host.address
171-
tar_path = host_dir / "tmp" / "metrics.tar.gz"
172-
173-
if tar_path.exists():
174-
with tarfile.open(tar_path, "r:gz") as tar:
175-
tar.extractall(path=host_dir)
176-
177-
tar_path.unlink()
178-
try:
179-
(host_dir / "tmp").rmdir()
180-
except OSError:
181-
pass
182-
# Clean up all containers for next run
183-
with en.actions(roles=roles) as a:
184-
a.shell("docker rm -f $(docker ps -aq) || true")
70+
run_label = f"run{run_idx}"
71+
print(f"\n>>> Starting {run_label}...")
72+
73+
# Ensure a fresh metrics directory on every node and push component files
74+
with en.actions(roles=roles) as a:
75+
a.file(path="/tmp/metrics", state="absent")
76+
a.file(path="/tmp/metrics", state="directory", mode="0777")
77+
a.file(path=COMPONENTS_PATH, state="directory")
78+
a.copy(src="config/redis/components-grid5000/",
79+
dest=COMPONENTS_PATH + "/")
80+
81+
# Deploy Containers on Arbitrator
82+
with en.actions(roles=roles["arbitrator"]) as a:
83+
a.docker_container(
84+
name="redis", image=REDIS_IMAGE,
85+
network_mode="host", state="started"
86+
)
87+
a.docker_container(
88+
name="placement", image=PLACEMENT_IMAGE,
89+
network_mode="host", state="started",
90+
command=["./placement", "--port", "50006"]
91+
)
92+
a.docker_container(
93+
name="arbitrator-sidecar", image=SIDECAR_IMAGE,
94+
network_mode="host", state="started",
95+
command=[
96+
"./daprd",
97+
"--app-id", "arbitrator-sidecar",
98+
"--app-port", "3000",
99+
"--resources-path", "/components",
100+
"--placement-host-address", "localhost:50006",
101+
"--metrics-port", "9091",
102+
],
103+
volumes=[f"{COMPONENTS_PATH}:/components"],
104+
)
105+
a.docker_container(
106+
name="arbitrator", image=ACTOR_IMAGE,
107+
network_mode="host", state="started",
108+
volumes=["/tmp/metrics:/metrics:rw"],
109+
env={
110+
"ROLE": "arbitrator",
111+
"NUMBER_OF_PHILOSOPHERS": "6",
112+
"DAPR_HTTP_ENDPOINT": "http://localhost:3500",
113+
"DAPR_GRPC_ENDPOINT": "http://localhost:50001",
114+
"METRICS_DIRECTORY": "/metrics",
115+
},
116+
)
117+
118+
# Deploy Containers on Workers
119+
for i, host in enumerate(roles["worker"]):
120+
with en.actions(pattern_hosts=host.address, roles=roles) as a:
121+
a.docker_container(
122+
name="redis", image=REDIS_IMAGE,
123+
network_mode="host", state="started"
124+
)
125+
a.docker_container(
126+
name="placement", image=PLACEMENT_IMAGE,
127+
network_mode="host", state="started",
128+
command=["./placement", "--port", "50006"]
129+
)
130+
a.docker_container(
131+
name=f"w{i}-sidecar", image=SIDECAR_IMAGE,
132+
network_mode="host", state="started",
133+
command=[
134+
"./daprd",
135+
"--app-id", f"w{i}-sidecar",
136+
"--app-port", "3000",
137+
"--resources-path", "/components",
138+
"--placement-host-address", "localhost:50006",
139+
"--metrics-port", "9091",
140+
],
141+
volumes=[f"{COMPONENTS_PATH}:/components"],
142+
)
143+
a.docker_container(
144+
name=f"w{i}", image=ACTOR_IMAGE,
145+
network_mode="host", state="started",
146+
volumes=["/tmp/metrics:/metrics:rw"],
147+
env={
148+
"PHILOSOPHER_ID": str(i),
149+
"DAPR_HTTP_ENDPOINT": "http://localhost:3500",
150+
"DAPR_GRPC_ENDPOINT": "http://localhost:50001",
151+
"METRICS_DIRECTORY": "/metrics",
152+
},
153+
)
154+
155+
# Wait for data collection
156+
print(f"--- {run_label}: Collecting data for {TIME_BEFORE_FETCH}s ---")
157+
for _ in tqdm(range(TIME_BEFORE_FETCH), desc=run_label, unit="s",
158+
mininterval=60):
159+
time.sleep(1)
160+
161+
# Fetch and Organize Results locally into run1, run2, etc.
162+
run_dest = LOCAL_ROOT / run_label
163+
run_dest.mkdir(parents=True, exist_ok=True)
164+
165+
with en.actions(roles=roles) as a:
166+
a.archive(path="/tmp/metrics", dest="/tmp/metrics.tar.gz", format="gz")
167+
a.fetch(src="/tmp/metrics.tar.gz", dest=str(run_dest), flat=False)
168+
169+
# Local extraction
170+
print(f"--- {run_label}: Extracting results ---")
171+
for host in en.get_hosts(roles):
172+
host_dir = run_dest / host.address
173+
tar_path = host_dir / "tmp" / "metrics.tar.gz"
174+
175+
if tar_path.exists():
176+
with tarfile.open(tar_path, "r:gz") as tar:
177+
tar.extractall(path=host_dir)
178+
179+
tar_path.unlink()
180+
try:
181+
(host_dir / "tmp").rmdir()
182+
except OSError:
183+
pass
184+
185+
# Clean up all containers for next run
186+
with en.actions(roles=roles) as a:
187+
a.shell("docker rm -f $(docker ps -aq) || true")
185188

186189
print(f"\n--- SUCCESS: All {NUM_RUNS} runs finished. Data in {LOCAL_ROOT} ---")
187-
provider.destroy()
190+
provider.destroy()

0 commit comments

Comments
 (0)