[Kube Runtime] Refine init and runtime scripts in k8s pods.

src/kube-runtime/build/kube-runtime.dockerfile

@@ -23,4 +23,4 @@ WORKDIR /pai-runtime
 COPY --from=frameworkcontroller/frameworkbarrier:v0.3.0 $BARRIER_DIR/frameworkbarrier .
 COPY src/ ./
 
-CMD ["/bin/sh", "-c", "/pai-runtime/entry"]
+CMD ["/bin/sh", "-c", "/pai-runtime/init"]

src/kube-runtime/src/entry → src/kube-runtime/src/init

@@ -18,24 +18,24 @@
 # OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 
 
-# This script is entrypoint of pai runtime, this will copy run to /usr/local/pai/run
-# which will be the entrypoint of task container. So `entry` running in runtime
-# container, `/usr/local/pai/run` running in task container, and will be pid 1 in
-# task container
+# This init script will be executed inside init container,
+# all scripts under init.d will be executed in priority order.
+# Init scripts will do preparations before user container starts.
+# Runtime script will be executed as the entrypoint of user container
+# and will be pid 1 process in user container.
 
 PAI_DIR=/usr/local/pai
-INIT_DIR=${PAI_DIR}/init
+RUNTIME_DIR=${PAI_DIR}/runtime.d
 PAI_LOG_DIR=${PAI_DIR}/logs
 
-MAIN_SH=${INIT_DIR}/main.sh
 LOG_FILE=${PAI_LOG_DIR}/${FC_POD_UID}_entry.log
 
 mkdir -p $PAI_LOG_DIR
-cp /pai-runtime/run $PAI_DIR
-cp -r /pai-runtime/init $PAI_DIR
+cp /pai-runtime/runtime $PAI_DIR
+cp -r /pai-runtime/runtime.d $PAI_DIR
 
 # do preparation for each script
-for i in `find /pai-runtime/prep/ -type f -regex ".*.sh"` ; do
+for i in `find /pai-runtime/init.d/ -type f -regex ".*.sh"` ; do
   file_name=`basename $i`
   echo -e "##### [${i} start] ##### \n"
   $i >> ${PAI_LOG_DIR}/${FC_POD_UID}_prep.log 2>&1
@@ -47,14 +47,11 @@ cd $PAI_LOG_DIR
 
 /pai-runtime/frameworkbarrier > $PAI_LOG_DIR/${FC_POD_UID}_barrier.log 2>&1
 echo "barrier returns $?" >> $LOG_FILE
-python /pai-runtime/parse.py framework.json > $PAI_DIR/runtime_env.sh 2> $PAI_LOG_DIR/${FC_POD_UID}_parse.log
+python /pai-runtime/runtime.d/parse.py framework.json > $PAI_DIR/runtime_env.sh 2> $PAI_LOG_DIR/${FC_POD_UID}_parse.log
 echo "parser.py returns $?" >> $LOG_FILE
 
-# prepare main.sh, which is where user command get started
-echo '#!/bin/sh' >> $MAIN_SH
-echo -e "\n\n$USER_CMD" >> $MAIN_SH
-
-chmod +x $MAIN_SH
+# prepare user.sh, which is where user command get started
+echo -e "\n$USER_CMD" >> ${RUNTIME_DIR}/user.sh
 
 # debug
 echo -e "finished entry\nmain.sh has:" >> $LOG_FILE

src/kube-runtime/src/run → src/kube-runtime/src/runtime

@@ -18,32 +18,25 @@
 # OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 
 
-# This script tries to behave like initd, will execute shell scripts under
-# `/usr/local/pai/init`, main.sh will get a special treat: it will start after
-# all other script and if main.sh exit this initd will kill all other processes in
-# container.
+# This runtime script will be executed inside user container,
+# all scripts under runtime.d will be executed in priority order.
+# User's commands will start in the end, and whole runtime script
+# will exit after user's commands exit.
 
 PAI_DIR=/usr/local/pai
-INIT_DIR=${PAI_DIR}/init
+RUNTIME_DIR=${PAI_DIR}/runtime.d
 #PAI_LOG_DIR=/usr/local/pai/logs/attempt-${FC_FRAMEWORK_ATTEMPT_ID}/role-${FC_TASKROLE_NAME}/idx-${FC_TASK_INDEX}/attempt-${FC_TASK_ATTEMPT_ID}/
 PAI_LOG_DIR=${PAI_DIR}/logs
 
 . $PAI_DIR/runtime_env.sh
 
-for i in `find $INIT_DIR/ -type f -regex ".*.sh"` ; do
-  file_name=`basename $i`
-  if [ $file_name = "main.sh" ] ; then
-    echo "skip main.sh for now"
-    continue
-  else
-    echo "starting ${file_name}"
-    $i > ${PAI_LOG_DIR}/${FC_POD_UID}_${file_name}_init.log 2>&1 &
-  fi
-done
-
-echo "starting main.sh"
-$INIT_DIR/main.sh 2>&1 | tee ${PAI_LOG_DIR}/${FC_POD_UID}_main.log & # TODO tee may not exist in user's container
-MAIN_PID=$!
-
-echo "wait for main"
-wait $MAIN_PID
+# To run other runtime scripts in user container,
+# execute them here in order.
+# ${RUNTIME_DIR}/runtime.sh > ${PAI_LOG_DIR}/${FC_POD_UID}_runtime.log 2>&1 &
+
+printf "[INFO] USER COMMAND START\n"
+${RUNTIME_DIR}/user.sh 2>&1 | tee ${PAI_LOG_DIR}/${FC_POD_UID}_main.log & # TODO tee may not exist in user's container
+USER_PID=$!
+
+printf "[INFO] USER COMMAND END\n"
+wait ${USER_PID}

src/kube-runtime/src/prep/ssh.sh → src/kube-runtime/src/runtime.d/user.sh

@@ -17,4 +17,5 @@
 # DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
 # OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 
-# TODO prepare sshd binary to user
+
+# user's commands here