Merge pull request #3513 from vrothberg/analyses

Binary bloat analyses

Makefile

@@ -20,6 +20,7 @@ SHAREDIR_CONTAINERS ?= ${PREFIX}/share/containers
 ETCDIR ?= /etc
 TMPFILESDIR ?= ${PREFIX}/lib/tmpfiles.d
 SYSTEMDDIR ?= ${PREFIX}/lib/systemd/system
+BUILDFLAGS ?=
 BUILDTAGS ?= \
 	$(shell hack/apparmor_tag.sh) \
 	$(shell hack/btrfs_installed_tag.sh) \
@@ -147,10 +148,10 @@ test/goecho/goecho: .gopathok $(wildcard test/goecho/*.go)
 	$(GO) build -ldflags '$(LDFLAGS)' -o $@ $(PROJECT)/test/goecho
 
 podman: .gopathok $(PODMAN_VARLINK_DEPENDENCIES) ## Build with podman
-	$(GO) build -gcflags '$(GCFLAGS)' -asmflags '$(ASMFLAGS)' -ldflags '$(LDFLAGS_PODMAN)' -tags "$(BUILDTAGS)" -o bin/$@ $(PROJECT)/cmd/podman
+	$(GO) build $(BUILDFLAGS) -gcflags '$(GCFLAGS)' -asmflags '$(ASMFLAGS)' -ldflags '$(LDFLAGS_PODMAN)' -tags "$(BUILDTAGS)" -o bin/$@ $(PROJECT)/cmd/podman
 
 podman-remote: .gopathok $(PODMAN_VARLINK_DEPENDENCIES) ## Build with podman on remote environment
-	$(GO) build -gcflags '$(GCFLAGS)' -asmflags '$(ASMFLAGS)' -ldflags '$(LDFLAGS_PODMAN)' -tags "$(BUILDTAGS) remoteclient" -o bin/$@ $(PROJECT)/cmd/podman
+	$(GO) build $(BUILDFLAGS) -gcflags '$(GCFLAGS)' -asmflags '$(ASMFLAGS)' -ldflags '$(LDFLAGS_PODMAN)' -tags "$(BUILDTAGS) remoteclient" -o bin/$@ $(PROJECT)/cmd/podman
 
 podman-remote-darwin: .gopathok $(PODMAN_VARLINK_DEPENDENCIES) ## Build with podman on remote OSX environment
 	CGO_ENABLED=0 GOOS=darwin $(GO) build -gcflags '$(GCFLAGS)' -asmflags '$(ASMFLAGS)' -ldflags '$(LDFLAGS_PODMAN)' -tags "remoteclient containers_image_openpgp exclude_graphdriver_devicemapper" -o bin/$@ $(PROJECT)/cmd/podman

dependencies/analyses/README.md

@@ -0,0 +1,88 @@
+# A set of scripts and instructions that help to analyze and debloat go-lang dependencies
+
+Note that all scripts mentioned below follow the [KISS principle](https://en.wikipedia.org/wiki/KISS_principle) on purpose.
+The scripts are meant to be used in combination to aid in understanding the packages' dependencies and how they contribute to the size of the compiled binary.
+
+## Size of packages
+
+To analyze the size of all go packages used during the build process, pass the `-work -a` build flags to `go build`.
+The `-a` flag forces go to rebuild all packages even if they are already up-to-date (e.g., in the build cache), while the `-work` flag instructs go to print the temporary work directory used for compiling the packages.
+The path to the temporary work directory of `go-build` must be passed to `go-archive-analysis.sh` by setting it as an environment variable.
+The analysis script will then read and parse the build data and print a sorted table of the package size in bytes followed by the package name.
+
+Running such an analysis on libpod may look as follows:
+
+```
+# 1) Build the podman binary with `-work -a`.
+[libpod]$ BUILDFLAGS="-work -a" make podman
+[...]
+WORK=/tmp/go-build794287815
+
+# 2) Set the work directory as an environment variable and call the analysis script
+[libpod]$ WORK=/tmp/go-build794287815 ./dependencies/analyses/go-archive-analysis.sh | head -n10
+17M github.com/containers/libpod/cmd/podman/cliconfig
+13M github.com/containers/libpod/vendor/github.com/DataDog/zstd
+10M github.com/containers/libpod/vendor/k8s.io/api/core/v1
+3.7M net/http
+3.7M github.com/containers/libpod/libpod
+3.2M runtime
+2.7M github.com/containers/libpod/vendor/github.com/gogo/protobuf/proto
+2.5M github.com/containers/libpod/vendor/k8s.io/apimachinery/pkg/apis/meta/v1
+2.3M github.com/containers/libpod/vendor/github.com/vishvananda/netlink
+2.1M github.com/containers/libpod/cmd/podman/varlink
+```
+
+The output of the `go-archive-analysis.sh` script is a sorted table with the size in bytes followed by the package.
+The size denotes the size of the compiled package (i.e., the `.a` file).
+
+
+## Size of symbols in binary
+
+Once the binary is compiled, we can run another set of analyses on it.
+The `nm-symbols-analysis.sh` is a wrapper around `go tool nm` and prints a table with the size in bytes followed by the symbol's name.
+To avoid information overload, the scripts prints only symbols from the text/code segment.
+
+Running such an analysis on libpod may look as follows:
+
+```
+# 1) Compile the binary
+[libpod]$ make podman
+[...]
+
+# 2) Run the script with the binary as an argument
+[libpod]$ ./dependencies/analyses/nm-symbols-analysis.sh ./bin/podman | grep "containers/libpod/libpod" | head -n10
+299             github.com/containers/libpod/libpod.(*BoltState).AddContainer
+658             github.com/containers/libpod/libpod.(*BoltState).AddContainerToPod
+2120            github.com/containers/libpod/libpod.(*BoltState).AddPod
+3773            github.com/containers/libpod/libpod.(*BoltState).AddPod.func1
+965             github.com/containers/libpod/libpod.(*BoltState).AddVolume
+1651            github.com/containers/libpod/libpod.(*BoltState).AddVolume.func1
+558             github.com/containers/libpod/libpod.(*BoltState).AllContainers
+282             github.com/containers/libpod/libpod.(*BoltState).AllContainers.func1
+1121            github.com/containers/libpod/libpod.(*BoltState).AllContainers.func1.1
+558             github.com/containers/libpod/libpod.(*BoltState).AllPods
+```
+
+Running the script can help identify sources of bloat and reveal potential candidates (e.g., entire packages, types, or function) for refactoring.
+
+
+## Dependency Tree
+
+Use the `dependency-tree.sh` script to figure out which package includes which packages.
+The output of the script has the format `package: dependency_1, dependency_2, ...`.
+Each line is followed by a blank line to make it easier to read.
+The script generates two files:
+
+ - `direct-tree.txt` - listing direct dependencies
+ - `transitive-tree.txt` - listing direct and transitive dependencies
+
+Running such a dependency-tree analysis may look as follows:
+
+
+```
+[libpod]$ ./dependencies/analyses/dependency-tree.sh github.com/containers/libpod
+[libpod]$ grep "^github.com/containers/libpod/pkg/registries" direct-tree.txt
+github.com/containers/libpod/pkg/registries: github.com/containers/libpod/vendor/github.com/containers/image/pkg/sysregistriesv2, github.com/containers/libpod/vendor/github.com/containers/image/types, github.com/containers/libpod/pkg/rootless, github.com/containers/libpod/vendor/github.com/docker/distribution/reference, github.com/containers/libpod/vendor/github.com/pkg/errors, os, path/filepath, strings
+```
+
+As shown above, the script's output can then be used to query for specific packages (e.g, with `grep`).

dependencies/analyses/dependency-tree.sh

@@ -0,0 +1,17 @@
+#!/usr/bin/bash
+
+if test "$#" -ne 1; then
+	echo "invalid arguments: usage: $0 path to package"
+	exit 1
+fi
+
+go list $1/... \
+	| xargs -d '\n' go list -f '{{ .ImportPath }}: {{ join .Imports ", " }}' \
+	| awk '{ printf "%s\n\n", $0 }' \
+	> direct-tree.tmp.$$ && mv -f direct-tree.tmp.$$ direct-tree.txt
+
+
+go list $1/... \
+	| xargs -d '\n' go list -f '{{ .ImportPath }}: {{ join .Deps ", " }}' \
+	| awk '{ printf "%s\n\n", $0 }' \
+	> transitive-tree.tmp.$$ && mv -f transitive-tree.tmp.$$ transitive-tree.txt

dependencies/analyses/go-archive-analysis.sh

@@ -0,0 +1,12 @@
+#!/usr/bin/bash
+
+if [ -z "$WORK" ]
+then
+	echo "WORK environment variable must be set"
+	exit 1
+fi
+
+grep --no-filename packagefile $WORK/**/importcfg \
+	| awk '{ split($2, data, "="); printf "%s ", data[1]; system("du -sh " data[2]) }' \
+	| awk '{ printf "%s %s\n", $2, $1 }' \
+	| sort -u | sort -rh

dependencies/analyses/nm-symbols-analysis.sh

@@ -0,0 +1,9 @@
+#!/usr/bin/bash
+
+if test "$#" -ne 1; then
+	echo "invalid arguments: usage: $0 path/to/binary"
+	exit 1
+fi
+
+go tool nm -size "$1" \
+	| awk 'NF==4 && $3=="t" {printf "%s\t\t%s\n", $2, $4}'