https://github.com/okuvshynov/flamereport
FlameGraph visualization for terminal
https://github.com/okuvshynov/flamereport
flamegraph profiling tui
Last synced: 5 months ago
JSON representation
FlameGraph visualization for terminal
- Host: GitHub
- URL: https://github.com/okuvshynov/flamereport
- Owner: okuvshynov
- License: mit
- Created: 2020-11-02T09:01:15.000Z (over 5 years ago)
- Default Branch: main
- Last Pushed: 2020-12-14T08:37:20.000Z (about 5 years ago)
- Last Synced: 2025-01-10T22:29:38.594Z (about 1 year ago)
- Topics: flamegraph, profiling, tui
- Language: Python
- Homepage:
- Size: 1.16 MB
- Stars: 9
- Watchers: 3
- Forks: 1
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Flame Graphs for terminal
This is an implementation of flame graphs right in the terminal.
For an overview of flame graphs and original implementation, check [Brendan Gregg's website](http://www.brendangregg.com/flamegraphs.html)
While not as powerful as GUI tools, having chart right in the terminal is more convenient in case of iterative process, like:
1. run a profile on remote machine
2. visualize and study the profile
3. make changes in code/workload/app-level settings/profiling settings/...
4. repeat
In specific case of Linux perf flamereport can be seen as an alternative to running 'perf report'.
## Flame graphs overview
Please check http://www.brendangregg.com/flamegraphs.html
## Usage example
### With provided sample
```$ python flame.py < samples/osx_dtrace```
### reproducing provided sample on Mac OS
1. Run dtrace with -q quiet flag:
```sudo dtrace -q -x stackframes=100 -n 'profile-99 /arg0/ { @[stack()] = count(); } tick-10s { exit(0); }' -o dtrace_stacks```
2. Get https://github.com/brendangregg/FlameGraph (needed for stack folding)
```$ git clone https://github.com/brendangregg/FlameGraph.git```
3. Now collapse stacks:
```$ ./FlameGraph/stackcollapse.pl < dtrace_stacks > stacks```
4. Run:
```$ python ./flametui/flame.py < stacks```
### With Linux perf
1. Get https://github.com/brendangregg/FlameGraph (needed for stack folding)
```$ git clone https://github.com/brendangregg/FlameGraph.git```
2. Do system-wide profile:
```$ perf record -F 999 -a -g -- sleep 10```
3. Collapse stacks:
```$ perf script | ./FlameGraph/stackcollapse-perf.pl > stacks```
4. View visualization
```$ python ./flametui/flame.py < ./stacks```
## Interactive commands
* left/h - select the block to the left
* right/l - select the block to the right
* up/k - select the block above
* down/j - select the block below
* f - focus. Zoom into selected frame; makes it fit whole width, hides the siblings.
* p - pin to top. This zooms into selected frame (like 'f') and hides parent frames. This is useful for deep traces, as there's no scrolling functionality.
* F - hard focus. Zoom into all frames with title of selected frame, pin them to top.
* x - eXclude frame. Removes the selected frame, its children and shrinks parents accordingly. This is useful when the graph is dominated by few large but not particularly interesting frames; It provides a more convenient view compared to focusing on 'smaller but more interesting' frames individually.
* r - reset focus. Resets the focus but keeps exculded frames excluded.
* R - hard reset, brings everything to default view; Useful after exclusions.
* single mouse click - select the block
* double mouse click - zoom into the block. Equivalent to 'f'
* / - search
* n/N - select next/prev block within the highlighted set of views
* q - quit
## Output description
Simple selection with auto-highlight matching frames
Selecting a cell with multiple frames in it. Pressing 'n' on previous example would bring us here:
In case 'x' was used to hide some of the less interesting frames, status will show how much was excluded:
Graph element examples
* '[foo---]' -- individual frame.
* '-' - also individual frame.
* '+' - aggregated frame view which contains several frames. Selecting it will show the details in the status area. Might have hidden descendents.
## Known limitations
* No color-coding for parts of stack traces (kernel/userspace/vm/native/etc.)