merge master and fix image tag to execute-3.6-gpu for gpu

.github/workflows/docker-publish.yaml

@@ -8,11 +8,11 @@ on:
 #  schedule:
 #    - cron: '41 9 * * *'
   push:
-    branches: [ master, gpu ]
+    branches: [ master, gpu, htcondor-10 ]
     # Publish semver tags as releases.
     tags: [ 'v*.*.*' ]
   pull_request:
-    branches: [ master, gpu ]
+    branches: [ master, gpu, htcondor-10 ]
 
 env:
   # Dockhub
@@ -86,6 +86,7 @@ jobs:
           tag_list=()
           tag_list+=(${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}:"latest")
           tag_list+=(${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}:$TIMESTAMP)
+          tag_list=("execute-3.6-gpu")
           # This causes the tag_list array to be comma-separated below,
           # which is required for build-push-action
           IFS=,

Dockerfile

@@ -1,4 +1,4 @@
-ARG BASE_IMAGE=opensciencegrid/software-base:3.6-el7-release
+ARG BASE_IMAGE=hub.opensciencegrid.org/opensciencegrid/software-base:3.6-el7-release
 FROM ${BASE_IMAGE}
 ARG BASE_IMAGE
 
@@ -47,34 +47,14 @@ RUN yum install --enablerepo=osg-upcoming -y condor
 
 RUN yum-config-manager --add-repo https://download.docker.com/linux/centos/docker-ce.repo
 RUN yum install -y docker-ce-cli
-RUN yum install -y http://mirror.grid.uchicago.edu/pub/mwt2/sw/el7/mwt2-sysview-worker-2.0.3-1.noarch.rpm
+RUN yum install -y http://mirror.grid.uchicago.edu/pub/mwt2/sw/el7/mwt2-sysview-worker-2.0.5-1.noarch.rpm
 RUN yum install -y python36-tabulate
 
-# Add CVMFSEXEC 
-#RUN git clone https://github.com/cvmfs/cvmfsexec /cvmfsexec \
-# && cd /cvmfsexec \
-# && ./makedist osg \
-# # /cvmfs-cache and /cvmfs-logs is where the cache and logs will go; possibly bind-mounted. \
-# # Needs to be 1777 so the unpriv user can use it. \
-# # (Can't just chown, don't know the UID of the unpriv user.) \
-# && mkdir -p /cvmfs-cache /cvmfs-logs \
-# && chmod 1777 /cvmfs-cache /cvmfs-logs \
-# && rm -rf dist/var/lib/cvmfs log \
-# && ln -s /cvmfs-cache dist/var/lib/cvmfs \
-# && ln -s /cvmfs-logs log \
-# # tar up and delete the contents of /cvmfsexec so the unpriv user can extract it and own the files. \
-# && tar -czf /cvmfsexec.tar.gz ./* \
-# && rm -rf ./* \
-# # Again, needs to be 1777 so the unpriv user can extract into it. \
-# && chmod 1777 /cvmfsexec
-
 COPY condor/*.conf /etc/condor/config.d/
 COPY cron/* /etc/cron.d/
 COPY supervisor/* /etc/supervisord.d/
 COPY image-config/* /etc/osg/image-config.d/
 COPY libexec/* /usr/local/libexec/
-COPY sysview-client/sysclient /bin/
-COPY sysview-client/client /usr/lib/python3.6/site-packages/sysview/client
 COPY scripts/condor_node_check.sh /usr/local/sbin/
 COPY scripts/entrypoint.sh /bin/entrypoint.sh
 

condor/01-ccb.conf

@@ -1,2 +1,3 @@
 CCB_ADDRESS = $(CONDOR_HOST)
 PRIVATE_NETWORK_NAME = $(UID_DOMAIN)
+TRUST_DOMAIN = head01.af.uchicago.edu

prometheus/exporter.py

@@ -2,6 +2,7 @@
 
 import os
 import time
+from subprocess import check_call
 from prometheus_client import start_http_server, Gauge, Enum
 import requests
 
@@ -12,7 +13,7 @@ class AppMetrics:
     application metrics into Prometheus metrics.
     """
 
-    def __init__(self, paths=None, app_port=80, polling_interval_seconds=5):
+    def __init__(self, paths=None, app_port=80, polling_interval_seconds=30):
         self.app_port = app_port
         self.polling_interval_seconds = polling_interval_seconds
         if not paths:  
@@ -55,10 +56,17 @@ def fetch(self):
         #self.total_uptime.set(status_data["total_uptime"])
 
         for p in self.paths:
-            if os.path.exists(p):
-                self.health.labels(path=p).state("healthy")
-            else:
+            try:
+                check_call(['test', '-e', p], timeout=10)
+            except:
                 self.health.labels(path=p).state("unhealthy")
+            else:
+                self.health.labels(path=p).state("healthy")
+            # thread could stuck in D wait and result will be stale
+            #if os.path.exists(p):
+            #    self.health.labels(path=p).state("healthy")
+            #else:
+            #    self.health.labels(path=p).state("unhealthy")
 
 def main():
     """Main entry point"""