Feature/cgroups v2 (#23)

Feature/cgroups v2 (#23)

* add support for "cpu" and "memory" cgroup controllers v2
* bump major version of cmonitor_collector and cmonitor_tools from 1.8-5 to 2.0-0
* add new data collected from kernel of Fedora35
* backward incompat change: change cgroups v1 memory stats replacing the "total_" prefix with "stat." to make them more similar to cgroups v2 stats
* add major version check in cmonitor_chart and cmonitor_statistics to avoid troubles when mixing by mistake recordings obtained with version 1.8-5 and version 2.0-0
* regenerate all examples

Author: f18m

.gitignore

@@ -27,3 +27,6 @@ collector/src/tests/centos7-Linux-3.10.0-x86_64/sample*/sample-timestamp
 collector/src/tests/ubuntu20.04-Linux-5.4.0-x86_64/result*.json
 collector/src/tests/ubuntu20.04-Linux-5.4.0-x86_64/sample*
 collector/src/tests/ubuntu20.04-Linux-5.4.0-x86_64/current-sample
+collector/src/tests/fedora35-Linux-5.14.17-x86_64/result*.json
+collector/src/tests/fedora35-Linux-5.14.17-x86_64/sample*
+collector/src/tests/fedora35-Linux-5.14.17-x86_64/current-sample

Constants.mk

@@ -12,8 +12,8 @@ RPM_TARBALL_DIR:=/tmp/cmonitor/tarball
 # IMPORTANT: other places where the version must be updated:
 #  - debian/changelog             -> to release a new Ubuntu package
 # See also https://github.com/f18m/cmonitor/wiki/new-release
-CMONITOR_VERSION:=1.8
-CMONITOR_RELEASE:=5
+CMONITOR_VERSION:=2.0
+CMONITOR_RELEASE:=0
 
 ifeq ($(DOCKER_LATEST),1)
 DOCKER_TAG=latest

README.md

@@ -95,20 +95,17 @@ with per-thread granularity.
 
 ## Supported Linux Kernels
 
-This project currently supports only cgroups v1. Work is in progress to support cgroups v2.
-However for the time being it means you can use `cmonitor-collector` utility on:
-* Fedora 30 and earlier releases
-* Ubuntu 21.04 and earlier releases
-* All other Linux distributions where the following command returns an empty response:
-
-```
-grep cgroup2 /proc/filesystems
-```
+Since version 2.0 this project supports both cgroups v1 and cgroups v2.
+This means that the "cmonitor_collector" utility can run on any Linux kernel regardless of its version and its boot options 
+(since boot options may alter the cgroups technology in use).
 
 Note that `cmonitor-collector` utility is currently unit-tested against:
-* Centos 7 Linux 3.10.0 kernel
+* Centos 7 (Linux kernel v3.10.0)
+* Ubuntu 20.04 (Linux kernel v5.4.0)
+* Fedora 35 (Linux kernel v5.14.17)
 
 Other kernels will be tested in near future. Of course pull requests are welcome to extend coverage.
+See [tests folder](collector/src/tests) for more details.
 
 ## How to install
 
@@ -231,21 +228,28 @@ Example results:
 #### Monitoring a Docker container launching cmonitor_collector on the baremetal
 
 In this case you can simply install cmonitor as RPM or APT package following instructions in [How to install](#section-id-65)
-and then launch the cmonitor_collector utility as any other Linux daemon, specifying the name of the cgroup associated with
+and then launch the `cmonitor_collector` utility as any other Linux daemon, specifying the name of the cgroup associated with
 the docker container to monitor.
+Finding out the cgroup associated with a Docker container can require some detailed information about your OS / runtime configuration;
+for example you should know whether you're using cgroups v1 or v2 and which Docker cgroup driver are you using (cgroupfs or systemd);
+please see the [official Docker page](https://docs.docker.com/config/containers/runmetrics/#find-the-cgroup-for-a-given-container)
+for more details.
 In the following example a [Redis](https://hub.docker.com/_/redis) docker container is launched with the name 'userapp' and its
-CPU, memory, network and disk usage are monitored launching a cmonitor_collector instance:
+CPU, memory, network and disk usage are monitored launching a `cmonitor_collector` instance:
 
 ```
 docker run --name userapp --detach redis:latest
 
 DOCKER_ID=$(docker ps -aq --no-trunc -f "name=userapp")
 
+CGROUP_NAME=docker/${DOCKER_ID}                       # when 'cgroupfs' driver is in use
+CGROUP_NAME=system.slice/docker-${DOCKER_ID}.scope    # when 'systemd' driver is in use
+
 # here we exploit the following fact: the cgroup of each Docker container 
-# is always named 'docker/container-ID'
+# is always named 'docker/container-ID'... at least when using Moby engine
 cmonitor_collector \
    --num-samples=until-cgroup-alive \
-   --cgroup-name=docker/${DOCKER_ID} \
+   --cgroup-name=${CGROUP_NAME} \
    --collect=cgroup_threads,cgroup_cpu,cgroup_memory --score-threshold=0 \
    --custom-metadata=cmonitor_chart_name:userapp \
    --sampling-interval=3 \
@@ -444,4 +448,4 @@ This fork supports only Linux x86_64 architectures; support for AIX/PowerPC (pre
 
 ## License
 
-Just like the [original project](http://nmon.sourceforge.net), this project is licensed under [GNU GPL 
+Just like the [original project](http://nmon.sourceforge.net), this project is licensed under [GNU GPL 2.0](LICENSE)

TODO.md

@@ -1,7 +1,15 @@
 ## TODO collector-side
 
-- Add 'blkio' cgroup data collection
-- Add benchmarks to see if fscanf() is actually slower compared to string2int() to optimize CPU usage
+- Add memory.pressure and cpu.pressure collection for cgroups v2
+- Start using "fmt-devel" to get very fast formatting support, see https://github.com/fmtlib/fmt
+  since the collector is mostly dealing with string/integer operations
+- Write Gbenchmark to compare 
+   * FILE* fopen() 
+   * open() syscall
+   * std::ifstream
+  since buffering is not interesting to our usecase... currently collector is using all of 3 solutions above
+  could not find any on the web!
+- Add 'io' cgroup v2 data collection
 - Add support for UDP data tx to InfluxDB
 
 ## TODO chart-side
@@ -18,7 +26,7 @@
    CMonitorHeaderInfo
 
 - add more sampled data for CMonitorCGroup, for several kernels
-- test more configurations for CMonitorCGroup
+- test more configurations for CMonitorCGroup, like dockers having no memory limit or no cpu limits
 - Test deployment on supported Linux distributions
 - Add LXC examples
 

collector/Makefile

@@ -129,7 +129,7 @@ cmonitor_musl: # build cmonitor inside a Docker from Alpine distro, to build a c
 	#        for docker builds, we strip them away to reduce docker size:
 	docker run --rm -it -v "${ROOT_DIR}":"/opt/src" -e "MUSL_BUILD=1" \
 		radupopescu/musl-builder \
-		sh -c "cd /opt/src/collector && gcc --version && make clean && make && make strip"
+		sh -c "cd /opt/src/collector && apk add gtest-dev && gcc --version && make clean && make -j && make strip"
 
 docker_image: cmonitor_musl
 	@cp -fv bin/musl/cmonitor_collector docker

collector/src/Makefile

@@ -5,6 +5,9 @@
 THIS_DIR:=$(shell dirname $(realpath $(lastword $(MAKEFILE_LIST))))
 ROOT_DIR:=$(shell readlink -f $(THIS_DIR)/../..)
 
+# IMPORTANT: we do require c++11, i.e. GCC 4.8.1 or higher;
+#            Centos7 which is one of the oldest-supported distributions provides gcc 4.8.5 so we
+#            cannot move to e.g. c++14/17/20 yet
 CXXFLAGS=-Wall -Werror -Wno-switch-bool -std=c++11 -DVERSION_STRING=\"$(RPM_VERSION)-$(RPM_RELEASE)\"
 
 ifeq ($(DEBUG),1)
@@ -35,7 +38,10 @@ MEMCHECK_COMMON_OPTS:=--tool=memcheck --track-origins=yes --malloc-fill=AF --fre
 
 
 OBJS = \
-    $(OUTDIR)/cgroups.o \
+    $(OUTDIR)/cgroups_config.o \
+	$(OUTDIR)/cgroups_cpuacct.o \
+	$(OUTDIR)/cgroups_memory.o \
+	$(OUTDIR)/cgroups_processes.o \
     $(OUTDIR)/header_info.o \
     $(OUTDIR)/logger.o \
     $(OUTDIR)/main.o \