commit | e6338cec201505cca5ca66c6457a155d40b6c95b | [log] [tgz] |
---|---|---|
author | Sanjay Ghemawat <sanjay@google.com> | Tue Aug 16 08:05:59 2022 -0700 |
committer | GitHub <noreply@github.com> | Tue Aug 16 08:05:59 2022 -0700 |
tree | 38b959188d648ddd15fc6438906cc5c428023a22 | |
parent | a41b82acbcb1f6a6d0998b0fddeb1f1adb7d8fc1 [diff] |
Added alternative flamegraph implementation that can show callers. (#716) Add an experimental flame-graph implementation. It can be selected in pprof's web interface using the new "Flame (experimental)" menu entry. At some point this new implementation may become the default. The new view is similar to flame-graph view. But it can show caller information as well. This should allow it to satisfy many users of Graph and Peek views as well. Let's illustrate with an example. Suppose we have profile data that consists of the following stacks: ``` 1000 main -> foo -> malloc 2000 main -> bar -> malloc ``` When main is selected, both the old and new views show: ``` [-------------------3000 main---------------------] [---1000 foo-----] [----------2000 bar------------] [---1000 malloc--] [----------2000 malloc---------] ``` But suppose now the user selects the right-most malloc slot. The old view will show just the path leading to that malloc: ``` [----------2000 main-----------] [----------2000 bar------------] [----------2000 malloc---------] ``` The new view will however show a flame-graph view that grows upwards that displays the call stacks leading to malloc: ``` [---1000 main----] [----------2000 main-----------] [---1000 foo-----] [----------2000 bar------------] [-------------------3000 malloc-------------------] ``` This caller display is useful when trying to determine expensive callers of function. A list of important differences between the new view and flame graphs: New view pros: 1. Callers are shown, e.g., all paths leading to malloc. 2. Shows self-cost clearly with a different saturation. 3. Font size is adjusted to fit more text into boxes. 4. Highlighting on hover shows other occurrences of a function. 5. Search works more like other views. 6. Pivot changes are reflected in browser history (so back and forward buttons can be used to navigate between different selections). 7. Allows eventual removal of the D3 dependency, which may make integrations into various environments easier. 8. Colors provide higher contrast between foreground and background. New view cons: 1. There are small differences in how things look and feel. 2. Color-scheme is very different. 3. Change triggered by selecting a new entry is not animated.
pprof is a tool for visualization and analysis of profiling data.
pprof reads a collection of profiling samples in profile.proto format and generates reports to visualize and help analyze the data. It can generate both text and graphical reports (through the use of the dot visualization package).
profile.proto is a protocol buffer that describes a set of callstacks and symbolization information. A common usage is to represent a set of sampled callstacks from statistical profiling. The format is described on the proto/profile.proto file. For details on protocol buffers, see https://developers.google.com/protocol-buffers
Profiles can be read from a local file, or over http. Multiple profiles of the same type can be aggregated or compared.
If the profile samples contain machine addresses, pprof can symbolize them through the use of the native binutils tools (addr2line and nm).
This is not an official Google product.
Prerequisites:
Go development kit of a supported version. Follow these instructions to prepare the environment.
Graphviz: http://www.graphviz.org/ Optional, used to generate graphic visualizations of profiles
To build and install it:
go install github.com/google/pprof@latest
The binary will be installed $GOPATH/bin
($HOME/go/bin
by default).
pprof can read a profile from a file or directly from a server via http. Specify the profile input(s) in the command line, and use options to indicate how to format the report.
% pprof -top [main_binary] profile.pb.gz Where main_binary: Local path to the main program binary, to enable symbolization profile.pb.gz: Local path to the profile in a compressed protobuf, or URL to the http service that serves a profile.
pprof -web [main_binary] profile.pb.gz
If no output formatting option is specified, pprof runs on interactive mode, where reads the profile and accepts interactive commands for visualization and refinement of the profile.
pprof [main_binary] profile.pb.gz This will open a simple shell that takes pprof commands to generate reports. Type 'help' for available commands/options.
If the -http
flag is specified, pprof starts a web server at the specified host:port that provides an interactive web-based interface to pprof. Host is optional, and is “localhost” by default. Port is optional, and is a random available port by default. -http=":"
starts a server locally at a random port.
pprof -http=[host]:[port] [main_binary] profile.pb.gz
The preceding command should automatically open your web browser at the right page; if not, you can manually visit the specified port in your web browser.
pprof can read perf.data
files generated by the Linux perf tool by using the perf_to_profile
program from the perf_data_converter package.
To view disassembly of profiles collected from Go programs compiled as Windows executables, the executable must be built with go build -buildmode=exe
. LLVM or GCC must be installed, so required tools like addr2line
and nm
are available to pprof
.
See doc/README.md for more detailed end-user documentation.
See CONTRIBUTING.md for contribution documentation.
See proto/README.md for a description of the profile.proto format.