fix compatibility issues with the latest version of apptainer

Makefile

@@ -1,4 +1,4 @@
-.PHONY: help rebuild build launch bash bashroot notebook warmup test singularity-build singularity-launch singularity-stop
+.PHONY: help rebuild build launch bash bashroot notebook warmup test apptainer-build apptainer-launch apptainer-shell apptainer-stop
 
 help:
 	@cat Makefile
@@ -8,6 +8,7 @@ ELEPHANT_WORKSPACE?=${PWD}/workspace
 ELEPHANT_IMAGE_NAME?=elephant-server:0.5.0
 ELEPHANT_NVIDIA_GID?=$$(ls -n /dev/nvidia0 2>/dev/null | awk '{print $$4}')
 ELEPHANT_DOCKER?=docker
+ELEPHANT_REDIS_PORT?=6379
 
 rebuild:
 	@IMAGEID=$$($(ELEPHANT_DOCKER) images -q $(ELEPHANT_IMAGE_NAME)); \
@@ -39,7 +40,7 @@ warmup:
 
 launch: warmup
 	$(ELEPHANT_DOCKER) run -it --rm $(GPU_ARG) --shm-size=8g -v $(ELEPHANT_WORKSPACE):/workspace -p 8080:80 -p 5672:5672 \
-	-e LOCAL_UID=$(shell id -u) -e LOCAL_GID=$(shell id -g) -e NVIDIA_GID=$(ELEPHANT_NVIDIA_GID) \
+	-e LOCAL_UID=$(shell id -u) -e LOCAL_GID=$(shell id -g) -e NVIDIA_GID=$(ELEPHANT_NVIDIA_GID) -e ELEPHANT_REDIS_PORT=$(ELEPHANT_REDIS_PORT)\
 	$(ELEPHANT_IMAGE_NAME)
 
 bash: warmup
@@ -60,17 +61,20 @@ test:
 	$(ELEPHANT_DOCKER) build -t $(ELEPHANT_IMAGE_NAME)-test -f Dockerfile-test . && $(ELEPHANT_DOCKER) image prune -f 
 	$(ELEPHANT_DOCKER) run -it --rm $(ELEPHANT_IMAGE_NAME)-test
 
-singularity-build:
-	singularity build --fakeroot elephant.sif elephant.def
-	singularity run --fakeroot elephant.sif
+apptainer-build:
+	apptainer build --fakeroot elephant.sif elephant.def
+	apptainer run --fakeroot --bind $(HOME):/root elephant.sif
 
-singularity-launch:
-	singularity instance start --nv --bind $(HOME)/.elephant_binds/var/lib:/var/lib,$(HOME)/.elephant_binds/var/log:/var/log,$(HOME)/.elephant_binds/var/run:/var/run,$(ELEPHANT_WORKSPACE):/workspace elephant.sif elephant
+apptainer-launch:
+	apptainer instance start --nv --bind $(HOME),$(HOME)/.elephant_binds/var/lib:/var/lib,$(HOME)/.elephant_binds/var/log:/var/log,$(HOME)/.elephant_binds/var/run:/var/run,$(HOME)/.elephant_binds/etc/nginx:/etc/nginx,$(HOME)/.elephant_binds/etc/rabbitmq:/etc/rabbitmq,$(ELEPHANT_WORKSPACE):/workspace elephant.sif elephant
 	if [ $(ELEPHANT_GPU) = all ]; then \
-		singularity exec instance://elephant /start.sh; \
+		apptainer exec --env ELEPHANT_REDIS_PORT=$(ELEPHANT_REDIS_PORT) instance://elephant /start.sh; \
 	else \
-		SINGULARITYENV_CUDA_VISIBLE_DEVICES=$(ELEPHANT_GPU) singularity exec instance://elephant /start.sh; \
+		apptainer exec --env CUDA_VISIBLE_DEVICES=$(ELEPHANT_GPU),ELEPHANT_REDIS_PORT=$(ELEPHANT_REDIS_PORT) instance://elephant /start.sh; \
 	fi 
 
-singularity-stop:
-	singularity instance stop elephant
+apptainer-shell:
+	apptainer shell --env ELEPHANT_REDIS_PORT=$(ELEPHANT_REDIS_PORT) --bind $(HOME),$(HOME)/.elephant_binds/var/lib:/var/lib,$(HOME)/.elephant_binds/var/log:/var/log,$(HOME)/.elephant_binds/var/run:/var/run,$(HOME)/.elephant_binds/etc/nginx:/etc/nginx,$(HOME)/.elephant_binds/etc/rabbitmq:/etc/rabbitmq elephant.sif
+
+apptainer-stop:
+	apptainer instance stop elephant

app/main.py

@@ -23,6 +23,8 @@
 # OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 # ==============================================================================
 """Flask endpoints."""
+import os
+
 from celery import Celery
 from flask import Flask
 from flask import request
@@ -31,7 +33,9 @@
 from apis import init_api
 from elephant.logging import logger
 from elephant.redis_util import TrainState
+from elephant.redis_util import REDIS_HOST
 from elephant.redis_util import REDIS_KEY_STATE
+from elephant.redis_util import REDIS_PORT
 
 
 def make_celery(app):
@@ -53,15 +57,13 @@ def __call__(self, *args, **kwargs):
 
 app = Flask(__name__)
 app.config.update(
-    broker_url='redis://localhost:6379',
-    result_backend='redis://localhost:6379',
+    broker_url=f'redis://{REDIS_HOST}:{REDIS_PORT}',
+    result_backend=f'redis://{REDIS_HOST}:{REDIS_PORT}',
     worker_redirect_stdouts=False,
     worker_prefetch_multiplier=1,
 )
 
 celery = make_celery(app)
-redis_client = FlaskRedis(app)
-redis_client.set(REDIS_KEY_STATE, TrainState.IDLE.value)
 
 
 init_api(app)

docker/rabbitmq-env.conf

@@ -0,0 +1 @@
+NODENAME=rabbit@localhost

docker/rabbitmq.sh

@@ -12,4 +12,4 @@ echo "*** Log in the WebUI at port 5672 (example: http:/localhost:5672) ***") &
 # $@ is used to pass arguments to the rabbitmq-server command.
 # For example if you use it like this: docker run -d rabbitmq arg1 arg2,
 # it will be as you run in the container rabbitmq-server arg1 arg2 
-rabbitmq-server $@
+/etc/init.d/rabbitmq-server $@

docker/supervisord.conf

@@ -22,7 +22,7 @@ stopsignal=QUIT
 priority=900
 
 [program:redis-server]
-command=redis-server
+command=redis-server --port %(ENV_ELEPHANT_REDIS_PORT)s
 stdout_logfile=/dev/stdout
 stdout_logfile_maxbytes=0
 stderr_logfile=/dev/stderr

docker/uwsgi.ini

@@ -6,8 +6,8 @@ cheaper = 2
 ; By default, when on demand, run up to 16 processes
 processes = 16
 socket = /tmp/uwsgi.sock
-chown-socket = nginx:nginx
-chmod-socket = 664
+# chown-socket = nginx:nginx
+# chmod-socket = 664
 ; Graceful shutdown on SIGTERM, see https://github.com/unbit/uwsgi/issues/849#issuecomment-118869386
 hook-master-start = unix_signal:15 gracefully_kill_them_all
 need-app = true

elephant-core/elephant/redis_util.py

@@ -25,6 +25,7 @@
 """Utils used for Redis."""
 
 from enum import Enum
+import os
 
 import redis
 
@@ -36,6 +37,8 @@
 REDIS_KEY_TIMEPOINT = 'timepoint'
 REDIS_KEY_UPDATE_ONGOING_SEG = 'update_ongoing_seg'
 REDIS_KEY_UPDATE_ONGOING_FLOW = 'update_ongoing_flow'
+REDIS_HOST = os.environ.get('ELEPHANT_REDIS_HOST', 'localhost')
+REDIS_PORT = os.environ.get('ELEPHANT_REDIS_PORT', '6379')
 
 
 class TrainState(Enum):
@@ -45,7 +48,8 @@ class TrainState(Enum):
 
 
 if RUN_ON_FLASK:
-    redis_client = redis.Redis.from_url('redis://localhost:6379/0')
+    redis_client = redis.Redis.from_url(f'redis://{REDIS_HOST}:{REDIS_PORT}/0')
+    redis_client.set(REDIS_KEY_STATE, TrainState.IDLE.value)
 else:
     redis_client = None
 

elephant.def

@@ -40,13 +40,14 @@ From: pytorch/pytorch:1.10.0-cuda11.3-cudnn8-runtime
 
     # Install and set up RabbbitMQ
     chmod +x /docker/install-rabbitmq.sh && /docker/install-rabbitmq.sh
+    cp /docker/rabbitmq-env.conf /etc/rabbitmq/rabbitmq-env.conf
 
     RABBITMQ_USER=user && RABBITMQ_PASSWORD=user
-    service rabbitmq-server start
+    /etc/init.d/rabbitmq-server start
     rabbitmqctl add_user $RABBITMQ_USER $RABBITMQ_PASSWORD 2>/dev/null
     rabbitmqctl set_user_tags $RABBITMQ_USER administrator
     rabbitmqctl set_permissions -p / $RABBITMQ_USER ".*" ".*" ".*"
-    service rabbitmq-server stop
+    /etc/init.d/rabbitmq-server stop
 
     # Set up nginx
     cp /docker/nginx.conf /etc/nginx/nginx.conf
@@ -80,16 +81,19 @@ From: pytorch/pytorch:1.10.0-cuda11.3-cudnn8-runtime
     mkdir -p $HOME/.elephant_binds/var/run
     cp -a /var/log $HOME/.elephant_binds/var/
     cp -a /var/lib $HOME/.elephant_binds/var/
+    mkdir -p $HOME/.elephant_binds/etc
+    cp -a /etc/nginx $HOME/.elephant_binds/etc/
+    cp -a /etc/rabbitmq $HOME/.elephant_binds/etc/
 
 %labels
     Author Ko Sugawara <[email protected]>
 
 %help
     This is a container to run ELEPHANt server.
     Usage:
-    singularity build --fakeroot elephant.sif elephant.def
-    singularity run --fakeroot elephant.sif
-    singularity instance start --nv --bind $HOME/.elephant_binds/var/lib:/var/lib,$HOME/.elephant_binds/var/log:/var/log,$HOME/.elephant_binds/var/run:/var/run,$ELEPHANT_WORKSPACE:/workspace elephant.sif elephant
-    singularity exec instance://elephant python /opt/elephant/script/dataset_generator.py --uint16 /workspace/datasets/elephant-demo/elephant-demo.h5 /workspace/datasets/elephant-demo
-    SINGULARITYENV_CUDA_VISIBLE_DEVICES=0 singularity exec instance://elephant /start.sh
-    singularity instance stop elephant
+    apptainer build --fakeroot elephant.sif elephant.def
+    apptainer run --bind $HOME elephant.sif
+    apptainer instance start --nv --bind $HOME,$HOME/.elephant_binds/var/lib:/var/lib,$HOME/.elephant_binds/var/log:/var/log,$HOME/.elephant_binds/var/run:/var/run,$HOME/.elephant_binds/etc/nginx:/etc/nginx,$HOME/.elephant_binds/etc/rabbitmq:/etc/rabbitmq,$ELEPHANT_WORKSPACE:/workspace elephant.sif elephant
+    apptainer exec instance://elephant python /opt/elephant/script/dataset_generator.py --uint16 /workspace/datasets/elephant-demo/elephant-demo.h5 /workspace/datasets/elephant-demo
+    SINGULARITYENV_CUDA_VISIBLE_DEVICES=0 apptainer exec instance://elephant /start.sh
+    apptainer instance stop elephant