Replace most shell script logic with Java (#85758)

Elasticsearch provides several command line tools, as well as the main script to start elasticsearch. While most of the logic is abstracted away for cli tools, the main elasticsearch script has hundreds of lines of platform specific shell code. That code is difficult to maintain because it uses many special shell features which then must also exist in other platforms (ie windows batch files). Additionally, the logic in these scripts are not easy to test, we must be on the actual platform and test with a full installation of Elasticsearch, which is relatively slow (compared to most in process tests). This commit replaces logic of the main server script, as well as the windows service management script, with Java. The new entrypoints use the CliToolLauncher. The server cli figures out all the jvm options and such necessary, then launches the real server process. If run in the foreground, the launcher will stay alive for the lifetime of Elasticsearch; the streams are effectively inherited so all output from Elasticsearch still goes to the console. If daemonizing, the launcher waits around until Elasticsearch is "ready" (this means the Node startup completed), then detaches and exits. Co-authored-by: William Brafford <[email protected]>
elastic · May 19, 2022 · b9c504b · b9c504b
1 parent 096d7fe
commit b9c504b
Show file tree

Hide file tree

Showing 58 changed files with 3,113 additions and 1,253 deletions.
diff --git a/distribution/build.gradle b/distribution/build.gradle
@@ -230,7 +230,7 @@ configure(subprojects.findAll { ['archives', 'packages'].contains(it.name) }) {
    *             Properties to expand when copying packaging files             *
    *****************************************************************************/
   configurations {
-    ['libs', 'libsVersionChecker', 'libsCliLauncher', 'libsServerCli', 'libsPluginCli', 'libsKeystoreCli', 'libsSecurityCli', 'libsGeoIpCli', 'libsAnsiConsole'].each {
+    ['libs', 'libsVersionChecker', 'libsCliLauncher', 'libsServerCli', 'libsWindowsServiceCli', 'libsPluginCli', 'libsKeystoreCli', 'libsSecurityCli', 'libsGeoIpCli', 'libsAnsiConsole'].each {
       create(it) {
         canBeConsumed = false
         canBeResolved = true
@@ -253,6 +253,7 @@ configure(subprojects.findAll { ['archives', 'packages'].contains(it.name) }) {
     libsVersionChecker project(':distribution:tools:java-version-checker')
     libsCliLauncher project(':distribution:tools:cli-launcher')
     libsServerCli project(':distribution:tools:server-cli')
+    libsWindowsServiceCli project(':distribution:tools:windows-service-cli')
     libsAnsiConsole project(':distribution:tools:ansi-console')
     libsPluginCli project(':distribution:tools:plugin-cli')
     libsKeystoreCli project(path: ':distribution:tools:keystore-cli')
@@ -278,6 +279,9 @@ configure(subprojects.findAll { ['archives', 'packages'].contains(it.name) }) {
         into('tools/server-cli') {
           from(configurations.libsServerCli)
         }
+        into('tools/windows-service-cli') {
+          from(configurations.libsWindowsServiceCli)
+        }
         into('tools/geoip-cli') {
           from(configurations.libsGeoIpCli)
         }
@@ -295,7 +299,6 @@ configure(subprojects.findAll { ['archives', 'packages'].contains(it.name) }) {
         }
       }
 
-
     modulesFiles = { platform ->
       copySpec {
         eachFile {

diff --git a/distribution/packages/src/common/systemd/elasticsearch.service b/distribution/packages/src/common/systemd/elasticsearch.service
@@ -6,6 +6,10 @@ After=network-online.target
 
 [Service]
 Type=notify
+# the elasticsearch process currently sends the notifications back to systemd
+# and for some reason exec does not work (even though it is a child). We should change
+# this notify access back to main (the default), see https://github.com/elastic/elasticsearch/issues/86475
+NotifyAccess=all
 RuntimeDirectory=elasticsearch
 PrivateTmp=true
 Environment=ES_HOME=/usr/share/elasticsearch

diff --git a/distribution/src/bin/elasticsearch b/distribution/src/bin/elasticsearch
@@ -1,139 +1,5 @@
 #!/bin/bash
 
-# CONTROLLING STARTUP:
-#
-# This script relies on a few environment variables to determine startup
-# behavior, those variables are:
-#
-#   ES_PATH_CONF -- Path to config directory
-#   ES_JAVA_OPTS -- External Java Opts on top of the defaults set
-#
-# Optionally, exact memory values can be set using the `ES_JAVA_OPTS`. Example
-# values are "512m", and "10g".
-#
-#   ES_JAVA_OPTS="-Xms8g -Xmx8g" ./bin/elasticsearch
-
-source "`dirname "$0"`"/elasticsearch-env
-
-CHECK_KEYSTORE=true
-ATTEMPT_SECURITY_AUTO_CONFIG=true
-DAEMONIZE=false
-ENROLL_TO_CLUSTER=false
-# Store original arg array as we will be shifting through it below
-ARG_LIST=("$@")
-
-while [ $# -gt 0 ]; do
-   if [[ $1 == "--enrollment-token" ]]; then
-     if [ $ENROLL_TO_CLUSTER = true ]; then
-       echo "Multiple --enrollment-token parameters are not allowed" 1>&2
-       exit 1
-     fi
-     ENROLL_TO_CLUSTER=true
-     ATTEMPT_SECURITY_AUTO_CONFIG=false
-     ENROLLMENT_TOKEN="$2"
-     shift
-   elif [[ $1 == "-h" || $1 == "--help" || $1 == "-V" || $1 == "--version" ]]; then
-     CHECK_KEYSTORE=false
-     ATTEMPT_SECURITY_AUTO_CONFIG=false
-   elif [[ $1 == "-d" || $1 == "--daemonize" ]]; then
-     DAEMONIZE=true
-   fi
-   if [[ $# -gt 0 ]]; then
-     shift
-   fi
-done
-
-if [ -z "$ES_TMPDIR" ]; then
-  ES_TMPDIR=`"$JAVA" -cp "$SERVER_CLI_CLASSPATH" org.elasticsearch.server.cli.TempDirectory`
-fi
-
-if [ -z "$LIBFFI_TMPDIR" ]; then
-  LIBFFI_TMPDIR="$ES_TMPDIR"
-  export LIBFFI_TMPDIR
-fi
-
-# get keystore password before setting java options to avoid
-# conflicting GC configurations for the keystore tools
-unset KEYSTORE_PASSWORD
-KEYSTORE_PASSWORD=
-if [[ $CHECK_KEYSTORE = true ]] \
-    && bin/elasticsearch-keystore has-passwd --silent
-then
-  if ! read -s -r -p "Elasticsearch keystore password: " KEYSTORE_PASSWORD ; then
-    echo "Failed to read keystore password on console" 1>&2
-    exit 1
-  fi
-fi
-
-if [[ $ENROLL_TO_CLUSTER = true ]]; then
-  CLI_NAME="auto-configure-node" \
-  CLI_LIBS="modules/x-pack-core,modules/x-pack-security,lib/tools/security-cli" \
-  bin/elasticsearch-cli "${ARG_LIST[@]}" <<<"$KEYSTORE_PASSWORD"
-elif [[ $ATTEMPT_SECURITY_AUTO_CONFIG = true ]]; then
-  # It is possible that an auto-conf failure prevents the node from starting, but this is only the exceptional case (exit code 1).
-  # Most likely an auto-conf failure will leave the configuration untouched (exit codes 73, 78 and 80), optionally printing a message
-  # if the error is uncommon or unexpected, but it should otherwise let the node to start as usual.
-  # It is passed in all the command line options in order to read the node settings ones (-E), while the other parameters are ignored
-  # (a small caveat is that it also inspects the -v option in order to provide more information on how auto config went)
-  if CLI_NAME="auto-configure-node" \
-    CLI_LIBS="modules/x-pack-core,modules/x-pack-security,lib/tools/security-cli" \
-    bin/elasticsearch-cli "${ARG_LIST[@]}" <<<"$KEYSTORE_PASSWORD"; then
-      :
-  else
-    retval=$?
-    # these exit codes cover the cases where auto-conf cannot run but the node should NOT be prevented from starting as usual
-    # eg the node is restarted, is already configured in an incompatible way, or the file system permissions do not allow it
-    if [[ $retval -ne 80 ]] && [[ $retval -ne 73 ]] && [[ $retval -ne 78 ]]; then
-      exit $retval
-    fi
-  fi
-fi
-
-# The JVM options parser produces the final JVM options to start Elasticsearch.
-# It does this by incorporating JVM options in the following way:
-#   - first, system JVM options are applied (these are hardcoded options in the
-#     parser)
-#   - second, JVM options are read from jvm.options and jvm.options.d/*.options
-#   - third, JVM options from ES_JAVA_OPTS are applied
-#   - fourth, ergonomic JVM options are applied
-ES_JAVA_OPTS=`export ES_TMPDIR; "$JAVA" -cp "$SERVER_CLI_CLASSPATH" -Des.distribution.type="$ES_DISTRIBUTION_TYPE" org.elasticsearch.server.cli.JvmOptionsParser "$ES_PATH_CONF" "$ES_HOME/plugins"`
-
-# Remove enrollment related parameters before passing the arg list to Elasticsearch
-for i in "${!ARG_LIST[@]}"; do
-  if [[ ${ARG_LIST[i]} = "--enrollment-token" || ${ARG_LIST[i]} = "$ENROLLMENT_TOKEN" ]]; then
-    unset 'ARG_LIST[i]'
-  fi
-done
-
-# manual parsing to find out, if process should be detached
-if [[ $DAEMONIZE = false ]]; then
-  exec \
-    "$JAVA" \
-    $ES_JAVA_OPTS \
-    -Des.path.home="$ES_HOME" \
-    -Des.path.conf="$ES_PATH_CONF" \
-    -Des.distribution.type="$ES_DISTRIBUTION_TYPE" \
-    -cp "$ES_CLASSPATH" \
-    org.elasticsearch.bootstrap.Elasticsearch \
-    "${ARG_LIST[@]}" <<<"$KEYSTORE_PASSWORD"
-else
-  exec \
-    "$JAVA" \
-    $ES_JAVA_OPTS \
-    -Des.path.home="$ES_HOME" \
-    -Des.path.conf="$ES_PATH_CONF" \
-    -Des.distribution.type="$ES_DISTRIBUTION_TYPE" \
-    -cp "$ES_CLASSPATH" \
-    org.elasticsearch.bootstrap.Elasticsearch \
-    "${ARG_LIST[@]}" \
-    <<<"$KEYSTORE_PASSWORD" &
-  retval=$?
-  pid=$!
-  [ $retval -eq 0 ] || exit $retval
-  if ! ps -p $pid > /dev/null ; then
-    exit 1
-  fi
-  exit 0
-fi
-
-exit $?
+CLI_NAME=server
+CLI_LIBS=lib/tools/server-cli
+source "`dirname "$0"`"/elasticsearch-cli
diff --git a/distribution/src/bin/elasticsearch-env b/distribution/src/bin/elasticsearch-env
@@ -32,10 +32,6 @@ while [ "`basename "$ES_HOME"`" != "bin" ]; do
 done
 ES_HOME=`dirname "$ES_HOME"`
 
-# now set the classpath
-ES_CLASSPATH="$ES_HOME/lib/*"
-SERVER_CLI_CLASSPATH="$ES_CLASSPATH:$ES_HOME/lib/tools/server-cli/*"
-
 # now set the path to java
 if [ ! -z "$ES_JAVA_HOME" ]; then
   JAVA="$ES_JAVA_HOME/bin/java"