Use a tiny base image for Docker builds

distribution/docker/build.gradle

@@ -58,7 +58,7 @@ ext.expansions = { Architecture architecture, boolean oss, DockerBase base, bool
 RUN curl --retry 8 -S -L \\
       --output /opt/elasticsearch.tar.gz \\
       https://artifacts.elastic.co/downloads/elasticsearch/$elasticsearch
-"""
+""".trim()
   }
 
   return [

distribution/docker/docker-test-entrypoint.sh

@@ -4,4 +4,4 @@ cd /usr/share/elasticsearch/bin/
 echo "testnode" > /tmp/password
 cat /tmp/password  | ./elasticsearch-keystore add -x -f -v 'xpack.security.transport.ssl.keystore.secure_password'
 cat /tmp/password  | ./elasticsearch-keystore add -x -f -v 'xpack.security.http.ssl.keystore.secure_password'
-/usr/local/bin/docker-entrypoint.sh | tee > /usr/share/elasticsearch/logs/console.log
+/usr/local/bin/docker-entrypoint.sh | tee /usr/share/elasticsearch/logs/console.log

distribution/docker/src/docker/Dockerfile

@@ -14,16 +14,17 @@
   layout/presentation here has been adjusted so that it looks reasonable when rendered,
   at the slight expense of how it looks here.
 */ %>
+<% if (docker_base == "ubi") { %>
 ################################################################################
 # Build stage 0 `builder`:
 # Extract Elasticsearch artifact
 ################################################################################
 
 FROM ${base_image} AS builder
-<% if (docker_base == 'ubi') { %>
+
 # Install required packages to extract the Elasticsearch distribution
 RUN ${package_manager} install -y tar gzip
-<% } %>
+
 # `tini` is a tiny but valid init for containers. This is used to cleanly
 # control how ES and any child processes are shut down.
 #
@@ -41,21 +42,167 @@ RUN set -eux ; \\
     curl --retry 8 -S -L -O https://github.com/krallin/tini/releases/download/v0.19.0/\${tini_bin}.sha256sum ; \\
     sha256sum -c \${tini_bin}.sha256sum ; \\
     rm \${tini_bin}.sha256sum ; \\
-    mv \${tini_bin} /tini ; \\
-    chmod +x /tini
+    mv \${tini_bin} /bin/tini ; \\
+    chmod +x /bin/tini
+<% } else { %>
+################################################################################
+# Stage 1. Build curl statically. Installing it from RPM on CentOS pulls in too
+# many dependencies.
+################################################################################
+FROM alpine:latest AS curl
+
+ENV VERSION 7.71.0
+ENV TARBALL_URL https://curl.haxx.se/download/curl-\${VERSION}.tar.xz
+ENV TARBALL_PATH curl-\${VERSION}.tar.xz
+
+# Install dependencies
+RUN apk add gnupg gcc make musl-dev openssl-dev openssl-libs-static file
+
+RUN mkdir /work
+WORKDIR /work
+
+# Fetch curl sources and files for validation
+RUN wget "https://daniel.haxx.se/mykey.asc" -O "curl-gpg.pub" && \\
+    wget "\${TARBALL_URL}.asc" -O "\${TARBALL_PATH}.asc" && \\
+    wget "\${TARBALL_URL}" -O "\${TARBALL_PATH}"
+
+# Validate source
+RUN gpg --import --always-trust "curl-gpg.pub" && \\
+    gpg --verify "\${TARBALL_PATH}.asc" "\${TARBALL_PATH}"
+
+# Unpack and build
+RUN tar xfJ "\${TARBALL_PATH}" && \\
+    cd "curl-\${VERSION}" && \\
+    ./configure --disable-shared --with-ca-fallback --with-ca-bundle=/etc/pki/tls/certs/ca-bundle.crt && \\
+    make curl_LDFLAGS="-all-static" && \\
+    cp src/curl /work/curl && \\
+    strip /work/curl
+
+################################################################################
+# Step 2. Create a minimal root filesystem directory. This will form the basis
+# for our image.
+################################################################################
+FROM ${base_image} AS rootfs
+
+ENV BUSYBOX_VERSION 1.31.0
+ENV TINI_VERSION 0.19.0
+
+# Start off with an up-to-date system
+RUN ${package_manager} update --setopt=tsflags=nodocs -y
+
+# Create a directory into which we will install files
+RUN mkdir /rootfs
+
+# Create required devices
+RUN mkdir -m 755 /rootfs/dev && \\
+    mknod -m 600 /rootfs/dev/console c 5 1 && \\
+    mknod -m 600 /rootfs/dev/initctl p && \\
+    mknod -m 666 /rootfs/dev/full c 1 7 && \\
+    mknod -m 666 /rootfs/dev/null c 1 3 && \\
+    mknod -m 666 /rootfs/dev/ptmx c 5 2 && \\
+    mknod -m 666 /rootfs/dev/random c 1 8 && \\
+    mknod -m 666 /rootfs/dev/tty c 5 0 && \\
+    mknod -m 666 /rootfs/dev/tty0 c 4 0 && \\
+    mknod -m 666 /rootfs/dev/urandom c 1 9 && \\
+    mknod -m 666 /rootfs/dev/zero c 1 5
+
+# Install a minimal set of dependencies, and some for Elasticsearch
+RUN ${package_manager} --installroot=/rootfs --releasever=/ --setopt=tsflags=nodocs \\
+      --setopt=group_package_types=mandatory -y  \\
+      --skip-broken \\
+      install basesystem bash zip zlib
+
+# `tini` is a tiny but valid init for containers. This is used to cleanly
+# control how ES and any child processes are shut down.
+#
+# The tini GitHub page gives instructions for verifying the binary using
+# gpg, but the keyservers are slow to return the key and this can fail the
+# build. Instead, we check the binary against the published checksum.
+#
+# Also, we use busybox instead of installing utility RPMs, which pulls in
+# all kinds of stuff we don't want.
+RUN set -e ; \\
+    TINI_BIN="" ; \\
+    BUSYBOX_ARCH="" ; \\
+    case "\$(arch)" in \\
+        aarch64) \\
+            BUSYBOX_ARCH='armv8l' ; \\
+            TINI_BIN='tini-arm64' ; \\
+            ;; \\
+        x86_64) \\
+            BUSYBOX_ARCH='x86_64' ; \\
+            TINI_BIN='tini-amd64' ; \\
+            ;; \\
+        *) echo >&2 "Unsupported architecture \$(arch)" ; exit 1 ;; \\
+    esac ; \\
+    curl --retry 8 -S -L -O "https://github.com/krallin/tini/releases/download/v0.19.0/\${TINI_BIN}" ; \\
+    curl --retry 8 -S -L -O "https://github.com/krallin/tini/releases/download/v0.19.0/\${TINI_BIN}.sha256sum" ; \\
+    sha256sum -c "\${TINI_BIN}.sha256sum" ; \\
+    rm "\${TINI_BIN}.sha256sum" ; \\
+    mv "\${TINI_BIN}" /rootfs/bin/tini ; \\
+    chmod +x /rootfs/bin/tini ; \\
+    curl --retry 10 -L -o /rootfs/bin/busybox \\
+      "https://busybox.net/downloads/binaries/\${BUSYBOX_VERSION}-defconfig-multiarch-musl/busybox-\${BUSYBOX_ARCH}" ; \\
+    chmod +x /rootfs/bin/busybox
+
+# Add links for most of the Busybox utilities
+RUN set -e ; \\
+    for path in \$( /rootfs/bin/busybox --list-full | grep -v bin/sh | grep -v telnet | grep -v unzip); do \\
+      ln /rootfs/bin/busybox /rootfs/\$path ; \\
+    done
+
+# Curl needs files under here. More importantly, we change Elasticsearch's
+# bundled JDK to use /etc/pki/ca-trust/extracted/java/cacerts instead of
+# the bundled cacerts.
+RUN mkdir -p /rootfs/etc && \\
+    cp -a /etc/pki /rootfs/etc/
+
+# Cleanup the filesystem
+RUN ${package_manager} --installroot=/rootfs -y clean all && \\
+    cd /rootfs && \\
+    rm -rf \\
+        etc/{X11,centos-release*,csh*,profile*,skel*,yum*} \\
+        sbin/sln \\
+        usr/bin/rpm \\
+        {usr,var}/games \\
+        usr/lib/{dracut,systemd,udev} \\
+        usr/lib64/X11 \\
+        usr/local \\
+        usr/share/{awk,centos-release,cracklib,desktop-directories,gcc-*,i18n,icons,licenses,xsessions,zoneinfo} \\
+        usr/share/{man,doc,info,games,gdb,ghostscript,gnome,groff,icons,pixmaps,sounds,backgrounds,themes,X11} \\
+        usr/{{lib,share}/locale,{lib,lib64}/gconv,bin/localedef,sbin/build-locale-archive} \\
+        var/cache/yum \\
+        var/lib/{rpm,yum} \\
+        var/log/yum.log
+
+# ldconfig
+RUN rm -rf /rootfs/etc/ld.so.cache /rootfs/var/cache/ldconfig && \\
+    mkdir -p --mode=0755 /rootfs/var/cache/ldconfig
+
+COPY --from=curl /work/curl /rootfs/usr/bin/curl
+
+################################################################################
+# Step 3. Fetch the Elasticsearch distribution and configure it for Docker
+################################################################################
+FROM ${base_image} AS builder
+<% } %>
 
 RUN mkdir /usr/share/elasticsearch
 WORKDIR /usr/share/elasticsearch
 
+# Fetch the appropriate Elasticsearch distribution for this architecture
 ${source_elasticsearch}
 
 RUN tar zxf /opt/elasticsearch.tar.gz --strip-components=1
+
+# Configure the distribution for Docker
 RUN sed -i -e 's/ES_DISTRIBUTION_TYPE=tar/ES_DISTRIBUTION_TYPE=docker/' /usr/share/elasticsearch/bin/elasticsearch-env
 RUN mkdir -p config config/jvm.options.d data logs
 RUN chmod 0775 config config/jvm.options.d data logs
 COPY config/elasticsearch.yml config/log4j2.properties config/
 RUN chmod 0660 config/elasticsearch.yml config/log4j2.properties
 
+<% if (docker_base == "ubi") { %>
 ################################################################################
 # Build stage 1 (the actual Elasticsearch image):
 #
@@ -65,12 +212,10 @@ RUN chmod 0660 config/elasticsearch.yml config/log4j2.properties
 
 FROM ${base_image}
 
-ENV ELASTIC_CONTAINER true
-
 RUN for iter in {1..10}; do \\
       ${package_manager} update --setopt=tsflags=nodocs -y && \\
       ${package_manager} install --setopt=tsflags=nodocs -y \\
-        nc shadow-utils zip unzip <%= docker_base == 'ubi' ? 'findutils procps-ng' : '' %> && \\
+        nc shadow-utils zip unzip findutils procps-ng && \\
       ${package_manager} clean all && exit_code=0 && break || exit_code=\$? && echo "${package_manager} error: retry \$iter in 10s" && \\
       sleep 10; \\
     done; \\
@@ -80,10 +225,30 @@ RUN groupadd -g 1000 elasticsearch && \\
     adduser -u 1000 -g 1000 -G 0 -d /usr/share/elasticsearch elasticsearch && \\
     chmod 0775 /usr/share/elasticsearch && \\
     chown -R 1000:0 /usr/share/elasticsearch
+<% } else { %>
+################################################################################
+# Stage 4. Build the final image, using the rootfs above as the basis, and
+# copying in the Elasticsearch distribution
+################################################################################
+FROM scratch
+
+# Setup the initial filesystem.
+COPY --from=rootfs /rootfs /
+
+RUN addgroup -g 1000 elasticsearch && \\
+    adduser -D -u 1000 -G elasticsearch -g elasticsearch -h /usr/share/elasticsearch elasticsearch && \\
+    addgroup elasticsearch root && \\
+    chmod 0775 /usr/share/elasticsearch && \\
+    chgrp 0 /usr/share/elasticsearch
+<% } %>
+
+ENV ELASTIC_CONTAINER true
 
 WORKDIR /usr/share/elasticsearch
 COPY --from=builder --chown=1000:0 /usr/share/elasticsearch /usr/share/elasticsearch
-COPY --from=builder --chown=0:0 /tini /tini
+<% if (docker_base == "ubi") { %>
+COPY --from=builder --chown=0:0 /bin/tini /bin/tini
+<% } %>
 
 # Replace OpenJDK's built-in CA certificate keystore with the one from the OS
 # vendor. The latter is superior in several ways.
@@ -94,14 +259,15 @@ ENV PATH /usr/share/elasticsearch/bin:\$PATH
 
 COPY bin/docker-entrypoint.sh /usr/local/bin/docker-entrypoint.sh
 
-# The JDK's directories' permissions don't allow `java` to be executed under a different
-# group to the default. Fix this.
+# 1. The JDK's directories' permissions don't allow `java` to be executed under a different
+#    group to the default. Fix this.
+# 2. Sync the user and group permissions of /etc/passwd
+# 3. Set correct permissions of the entrypoint
+# 4. Ensure that there are no files with setuid or setgid, in order to mitigate "stackclash" attacks.
 RUN find /usr/share/elasticsearch/jdk -type d -exec chmod 0755 '{}' \\; && \\
     chmod g=u /etc/passwd && \\
-    chmod 0775 /usr/local/bin/docker-entrypoint.sh
-
-# Ensure that there are no files with setuid or setgid, in order to mitigate "stackclash" attacks.
-RUN find / -xdev -perm -4000 -exec chmod ug-s {} +
+    chmod 0775 /usr/local/bin/docker-entrypoint.sh && \\
+    find / -xdev -perm -4000 -exec chmod ug-s {} +
 
 EXPOSE 9200 9300
 
@@ -138,7 +304,10 @@ RUN mkdir /licenses && \\
 <% } %>
 USER elasticsearch:root
 
-ENTRYPOINT ["/tini", "--", "/usr/local/bin/docker-entrypoint.sh"]
+# Our actual entrypoint is `tini`, a minimal but functional init program. It
+# calls the entrypoint we provide, while correctly forwarding signals.
+ENTRYPOINT ["/bin/tini", "--", "/usr/local/bin/docker-entrypoint.sh"]
+
 # Dummy overridable parameter parsed by entrypoint
 CMD ["eswrapper"]
 

docs/reference/setup/install/docker.asciidoc

@@ -1,8 +1,9 @@
 [[docker]]
 === Install {es} with Docker
 
-{es} is also available as Docker images.
-The images use https://hub.docker.com/_/centos/[centos:8] as the base image.
+{es} is also available as Docker images. Starting with version 8.0.0, these
+are based upon a tiny core of essential files. Prior versions used
+https://hub.docker.com/_/centos/[centos:8] as the base image.
 
 A list of all published Docker images and tags is available at
 https://www.docker.elastic.co[www.docker.elastic.co]. The source files
@@ -424,4 +425,23 @@ You must explicitly accept them either by:
 See {plugins}/_other_command_line_parameters.html[Plugin management]
 for more information.
 
+The {es} Docker image only includes what is required to run {es}, and does
+not provide a package manager. It is possible to add additional utilities
+with a multi-phase Docker build. You must also copy any dependencies, for
+example shared libraries.
+
+[source,sh,subs="attributes"]
+--------------------------------------------
+FROM centos:8 AS builder
+yum install -y some-package
+
+FROM docker.elastic.co/elasticsearch/elasticsearch:{version}
+COPY --from=builder /usr/bin/some-utility /usr/bin/
+COPY --from=builder /usr/lib/some-lib.so /usr/lib/
+--------------------------------------------
+
+You should use `centos:8` as a base in order to avoid incompatibilities.
+Use http://man7.org/linux/man-pages/man1/ldd.1.html[`ldd`] to list the
+shared libraries required by a utility.
+
 include::next-steps.asciidoc[]