Containerization of TM and TDS

agent/config/firewalld/pbench-dcgm-exporter.xml

@@ -0,0 +1,7 @@
+<?xml version="1.0" encoding="utf-8"?>
+<service>
+  <short>pbench-dcgm-exporter</short>
+  <description>Pbench Agent Prometheus dcgm-exporter</description>
+  <port protocol="tcp" port="9400"/>
+</service>
+

agent/config/firewalld/pbench-redis.xml

@@ -1,6 +1,6 @@
 <?xml version="1.0" encoding="utf-8"?>
 <service>
-  <short>pbench-tool-data-sink</short>
-  <description>Pbench Agent Tool Data Sink</description>
-  <port protocol="tcp" port="8080"/>
+  <short>pbench-redis</short>
+  <description>Pbench Agent Redis Server</description>
+  <port protocol="tcp" port="17001"/>
 </service>

agent/config/firewalld/pbench-tool-data-sink.xml

@@ -1,6 +1,6 @@
 <?xml version="1.0" encoding="utf-8"?>
 <service>
-  <short>pbench-redis</short>
-  <description>Pbench Agent Redis Server</description>
-  <port protocol="tcp" port="17001"/>
+  <short>pbench-tool-data-sink</short>
+  <description>Pbench Agent Tool Data Sink</description>
+  <port protocol="tcp" port="8080"/>
 </service>

agent/containers/images/Dockerfile.base.j2

@@ -13,9 +13,9 @@ RUN \
     {{ pkgmgr }} module -y disable python38 && \
 {% endif %}
 {% if distro_image.startswith('centos') %}
-    {{ pkgmgr }} install -y https://dl.fedoraproject.org/pub/epel/epel-release-latest-{{ distro_image.split(':', 1)[1] }}.noarch.rpm && \
+    {{ pkgmgr }} install -y --setopt=tsflags=nodocs https://dl.fedoraproject.org/pub/epel/epel-release-latest-{{ distro_image.split(':', 1)[1] }}.noarch.rpm && \
 {% endif %}
-    {{ pkgmgr }} install -y {% if distro_image == 'centos:8' %}--enablerepo powertools glibc-locale-source {% endif %} pbench-agent && \
+    {{ pkgmgr }} install -y --setopt=tsflags=nodocs {% if distro_image == 'centos:8' %}--enablerepo powertools glibc-locale-source {% endif %} pbench-agent && \
 {% if distro_image == 'centos:8' %}
     localedef -i en_US -f UTF-8 en_US.UTF-8 && \
 {% endif %}

agent/containers/images/Dockerfile.dcgmEX.j2

@@ -0,0 +1,15 @@
+# NOTE: Must be run with --privileged
+# RECOMMENDED: Use with the fedora image variants for direct compatibility
+FROM pbench-agent-tool-meister-{{ distro }}:{{ tag }}
+
+RUN {% if distro == 'centos-7' %}yum{% else %}dnf{% endif %} install -y 'dnf-command(config-manager)' && \
+    {% if distro == 'centos-7' %}yum{% else %}dnf{% endif %} config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/{{ distro.split("-")|join("") }}/x86_64/cuda-{{ distro.split("-")|join("") }}.repo && \
+    {% if distro == 'centos-7' %}yum{% else %}dnf{% endif %} clean expire-cache && \
+    {% if distro == 'centos-7' %}yum{% else %}dnf{% endif %} install -y nvidia-driver-cuda nvidia-modprobe datacenter-gpu-manager-2.1.4 golang && \
+    git clone https://github.com/NVIDIA/gpu-monitoring-tools.git && \
+    (cd gpu-monitoring-tools; git checkout tags/2.1.2 -b build; make binary install) && \
+    {% if distro == 'centos-7' %}yum{% else %}dnf{% endif %} -y clean all && \
+    rm -rf /var/cache/{% if distro == 'centos-7' %}yum{% else %}dnf{% endif %}
+
+ENV NVIDIA_DISABLE_REQUIRE="true" \
+    NVIDIA_VISIBLE_DEVICES=all

agent/containers/images/Dockerfile.layered.j2

@@ -1,10 +1,14 @@
 # {{ distro }} pbench-agent {{ kind }} image
 FROM pbench-agent-base-{{ distro }}:{{ tag }}
 
+{% if kind in ('tools', 'all') %}
+COPY ./{{ distro }}-pcp.repo /etc/yum.repos.d/pcp.repo
+{% endif %}
+
 # Install all the RPMs required for this image.
 #
 # FIXME: this is not exhaustive, it does not include RPMs to support
 #        Kubernetes or RHV environments.
-RUN {% if distro == 'centos-7' %}yum{% else %}dnf{% endif %} install -y {% if distro == 'centos-8' %}--enablerepo powertools {% endif %}{{ rpms }} && \
+RUN {% if distro == 'centos-7' %}yum{% else %}dnf{% endif %} install -y --setopt=tsflags=nodocs {% if distro == 'centos-8' %}--enablerepo powertools {% endif %}{% if kind in ('tools', 'all') %}--enablerepo pcp {% endif %}{{ rpms }} && \
     {% if distro == 'centos-7' %}yum{% else %}dnf{% endif %} -y clean all && \
     rm -rf /var/cache/{% if distro == 'centos-7' %}yum{% else %}dnf{% endif %}

agent/containers/images/Dockerfile.tds.j2

@@ -0,0 +1,9 @@
+# {{ distro }} pbench-agent-tool-data-sink  image
+FROM pbench-agent-tools-{{ distro }}:{{ tag }}
+
+VOLUME /var/lib/pbench-agent
+
+# Port 8080 should be Bottle server, 9090 optional Prometheus server, and 44566
+# the optional pmproxy server.
+EXPOSE 8080 9090 44566
+ENTRYPOINT [ "/opt/pbench-agent/util-scripts/tool-meister/tool-data-sink-ep" ]

agent/containers/images/Dockerfile.tm.j2

@@ -0,0 +1,7 @@
+# {{ distro }} pbench-agent-tool-meister image
+FROM pbench-agent-tools-{{ distro }}:{{ tag }}
+
+# Port 9400 should be the optional dcgm tool, 9100 the optional node_exporter
+# tool, and 55677 the pcp (pmcd) tool.
+EXPOSE 9100 9400 55677
+ENTRYPOINT [ "/opt/pbench-agent/util-scripts/tool-meister/tool-meister-ep" ]

agent/containers/images/Makefile

@@ -27,9 +27,43 @@ IMAGE_REPO = docker://quay.io/pbench
 # Not intended to be overridden with an environment variable.
 _REPO_TEMPLATE = ../../ansible/pbench/agent/roles/pbench_repo_install/templates/etc/yum.repos.d/pbench.repo.j2
 
+# NOTE: Currently we require 5.2.2 of the PCP RPMs because using the 5.2.3
+# version prevents us from integrating with Grafana, see PCP Issue #1183,
+# https://github.com/performancecopilot/pcp/issues/1183.
+# NOTE: We also have to enumerate so many RPMs because for CentOS 7 the RPM
+# dependency resolver does not properly resolve to the same version RPMs. Once
+# we no longer have to use v5.2.2, we can just list 3 RPMs: pcp-zeroconf,
+# pcp-system-tools, and pcp-gui.
+_PCP_RPMS = \
+	pcp-doc-5.2.2 \
+	pcp-gui-5.2.2 \
+	pcp-pmda-dm-5.2.2 \
+	pcp-pmda-nfsclient-5.2.2 \
+	pcp-pmda-openmetrics-5.2.2 \
+	pcp-system-tools-5.2.2 \
+	pcp-zeroconf-5.2.2 \
+	python3-pcp-5.2.2
 # The list of RPMs which provide the various tools we offer.
 # Not intended to be overridden with an environment variable.
-_TOOL_RPMS = prometheus2 node_exporter blktrace bpftrace cpupowerutils golang kernel-tools libvirt-client nmap-ncat numactl pbench-sysstat pcp-system-tools perf procps-ng strace tcpdump trace-cmd
+# Please keep the lists sorted.
+_TOOL_RPMS = \
+	blktrace \
+	bpftrace \
+	cpupowerutils \
+	golang \
+	kernel-tools \
+	libvirt-client \
+	nmap-ncat \
+	node_exporter \
+	numactl \
+	pbench-sysstat \
+	${_PCP_RPMS} \
+	perf \
+	procps-ng \
+	prometheus2 \
+	strace \
+	tcpdump \
+	trace-cmd
 
 # The list of RPMs for the default workloads we offer.
 # Not intended to be overridden with an environment variable.
@@ -41,8 +75,38 @@ _ALL_RPMS = ${_TOOL_RPMS} ${_WORKLOAD_RPMS}
 # By default we only build images for the following distributions:
 _DISTROS = centos-8 centos-7 fedora-33 fedora-32
 
+# By default we won't build the Tool Data Sink and Tool Meister images
 all: all-tags $(foreach distro, ${_DISTROS}, ${distro}-all-tagged)
 
+tds: all-tags $(foreach distro, ${_DISTROS}, ${distro}-tool-data-sink-tagged)
+
+tm: all-tags $(foreach distro, ${_DISTROS}, ${distro}-tool-meister-tagged)
+
+# We also offer targets per distribution target
+centos-8: all-tags centos-8-all-tagged
+
+centos-7: all-tags centos-7-all-tagged
+
+fedora-33: all-tags fedora-33-all-tagged
+
+fedora-32: all-tags fedora-32-all-tagged
+
+centos-8-tds: all-tags centos-8-tool-data-sink-tagged
+
+centos-7-tds: all-tags centos-7-tool-data-sink-tagged
+
+fedora-33-tds: all-tags fedora-33-tool-data-sink-tagged
+
+fedora-32-tds: all-tags fedora-32-tool-data-sink-tagged
+
+centos-8-tm: all-tags centos-8-tool-meister-tagged
+
+centos-7-tm: all-tags centos-7-tool-meister-tagged
+
+fedora-33-tm: all-tags fedora-33-tool-meister-tagged
+
+fedora-32-tm: all-tags fedora-32-tool-meister-tagged
+
 #+
 # Tagging targets
 #-
@@ -97,16 +161,34 @@ push-major-minor: $(foreach distro, ${_DISTROS}, ${distro}-push-major-minor)
 %-all-tagged: %-all %-tags.lis
 	./apply-tags pbench-agent-all-$* $*-tags.lis
 
-%-all: %-tools-tagged %-workloads-tagged %-all.Dockerfile
+%-all: %-workloads-tagged %-tool-data-sink-tagged %-tool-meister-tagged %-all.Dockerfile
 	./build-image all $* $*-tags.lis
 
 %-all.Dockerfile: Dockerfile.layered.j2 %-tags.lis
 	jinja2 Dockerfile.layered.j2 -D distro=$* -D tag="$$(grep -v -E '^v' $*-tags.lis)" -D kind="all" -D rpms="${_ALL_RPMS}" > ./$@
 
+%-tool-data-sink-tagged: %-tool-data-sink %-tags.lis
+	./apply-tags pbench-agent-tool-data-sink-$* $*-tags.lis
+
+%-tool-data-sink: %-tools-tagged %-tool-data-sink.Dockerfile
+	./build-image tool-data-sink $* $*-tags.lis
+
+%-tool-data-sink.Dockerfile: Dockerfile.tds.j2 %-tags.lis
+	jinja2 Dockerfile.tds.j2 -D distro=$* -D tag="$$(grep -v -E '^v' $*-tags.lis)" > ./$@
+
+%-tool-meister-tagged: %-tool-meister %-tags.lis
+	./apply-tags pbench-agent-tool-meister-$* $*-tags.lis
+
+%-tool-meister: %-tools-tagged %-tool-meister.Dockerfile
+	./build-image tool-meister $* $*-tags.lis
+
+%-tool-meister.Dockerfile: Dockerfile.tm.j2 %-tags.lis
+	jinja2 Dockerfile.tm.j2 -D distro=$* -D tag="$$(grep -v -E '^v' $*-tags.lis)" > ./$@
+
 %-tools-tagged: %-tools %-tags.lis
 	./apply-tags pbench-agent-tools-$* $*-tags.lis
 
-%-tools: %-base-tagged %-tools.Dockerfile
+%-tools: %-base-tagged %-tools.Dockerfile %-pcp.repo
 	./build-image tools $* $*-tags.lis
 
 %-tools.Dockerfile: Dockerfile.layered.j2 %-tags.lis
@@ -204,15 +286,23 @@ fedora-32-base.Dockerfile: Dockerfile.base.j2 fedora-32-pbench.repo
 # Helper target to build each distro's ".repo" and ".Dockerfile"
 all-dockerfiles: $(foreach distro, ${_DISTROS}, ${distro}-base.Dockerfile ${distro}-tools.Dockerfile ${distro}-workloads.Dockerfile ${distro}-all.Dockerfile)
 
-# Rule pattern dependencies on non-patterned targets have to be set up
-# separately for some reason.
-%.repo: ${_REPO_TEMPLATE}
+%-pbench.repo: %-pbench.yml ${_REPO_TEMPLATE}
+	jinja2 ${_REPO_TEMPLATE} $*-pbench.yml -o $@
+
+%-pbench.yml: repo.yml.j2
+	jinja2 repo.yml.j2 -D distro=$* -D url_prefix=${URL_PREFIX} -D test_suffix=${_TEST_SUFFIX} -D user=${USER} -o $@
+
+fedora-33-pcp.repo: pcp.repo.j2
+	jinja2 pcp.repo.j2 -D target=f33 -o $@
+
+fedora-32-pcp.repo: pcp.repo.j2
+	jinja2 pcp.repo.j2 -D target=f32 -o $@
 
-%.repo: %.yml
-	jinja2 ${_REPO_TEMPLATE} $*.yml -o $@
+centos-8-pcp.repo: pcp.repo.j2
+	jinja2 pcp.repo.j2 -D target=el8 -o $@
 
-%.yml: repo.yml.j2
-	jinja2 repo.yml.j2 -D distro=${@:-pbench.yml=} -D url_prefix=${URL_PREFIX} -D test_suffix=${_TEST_SUFFIX} -D user=${USER} -o $@
+centos-7-pcp.repo: pcp.repo.j2
+	jinja2 pcp.repo.j2 -D target=el7 -o $@
 
 clean:
 	rm -f *.Dockerfile *.repo *.yml *-tags.lis

agent/containers/images/pcp.repo.j2

@@ -0,0 +1,6 @@
+[pcp]
+name=pcp
+baseurl=https://dl.bintray.com/pcp/{{ target }}
+gpgcheck=0
+repo_gpgcheck=0
+enabled=1

agent/containers/images/push

@@ -27,7 +27,7 @@ function pushit {
     buildah push ${1} ${image_repo}/${1}
 }
 
-for image in base tools workloads all; do
+for image in base tools tool-meister tool-data-sink workloads all; do
     pushit pbench-agent-${image}-${distro}:${githash}
     pushit pbench-agent-${image}-${distro}:${ver}
     if [[ ! -z "${other}" ]]; then

agent/containers/images/tagit

@@ -16,6 +16,6 @@ function tagit {
     buildah tag ${1}:${githash} ${1}:${tag}
 }
 
-for image in base tools workloads all; do
+for image in base tools tool-meister tool-data-sink workloads all; do
     tagit pbench-agent-${image}-${distro}
 done