aboutsummaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorValentin Rothberg <rothberg@redhat.com>2019-07-08 14:04:32 +0200
committerValentin Rothberg <rothberg@redhat.com>2019-07-16 14:16:16 +0200
commitfb31cc95a596b9c7c09cbc712acd60027c32ddfa (patch)
treef1b55776d5767ad087a061fc7bece494b0700383
parent88058c3ce20bb88c2613b9b5ef9aa558bf8440b9 (diff)
downloadpodman-fb31cc95a596b9c7c09cbc712acd60027c32ddfa.tar.gz
podman-fb31cc95a596b9c7c09cbc712acd60027c32ddfa.tar.bz2
podman-fb31cc95a596b9c7c09cbc712acd60027c32ddfa.zip
analyses: add README.md
Signed-off-by: Valentin Rothberg <rothberg@redhat.com>
-rw-r--r--dependencies/analyses/README.md64
1 files changed, 64 insertions, 0 deletions
diff --git a/dependencies/analyses/README.md b/dependencies/analyses/README.md
new file mode 100644
index 000000000..70de3f621
--- /dev/null
+++ b/dependencies/analyses/README.md
@@ -0,0 +1,64 @@
+# A set of scripts and instructions that help to analyze and debloat go-lang dependencies.
+
+## 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 path of 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 size in bytes followed by the symbols 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/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.