Comments (3)
primiano
commented on May 31, 2024
On Sun, 21 Jan 2024 at 02:10, Jun Ye ***@***.***> wrote:
Thanks for the brilliant tools, We've been using in-process mode
successfully for a long time, but to address future needs, we want to merge
trace files from different processes. As there's no official Perfetto tool
for this, we're considering migrating to system mode. However, while
evaluating system mode, I have several concerns. First, the design doc
tells that traced will copy memory from shared to central when part of
shared memory is full, I think this potentially impacting CPU usage, we are
running tracing system for very long time, and the shared memory could
fulfilled in rapidly speed, since we only need to capture shared ring
buffer data to central buffer when issues occur, is there a configuration
in system mode to achieve this purpose?
The overhead of system mode vs in-process mode is essentially the same (%
process vs thread context switching). You have the same dataflow even in
in-process mode. There are little copies involved as tracing goes over a
shared memory buffer.
This should not be a concern.
Additionally, in in-process mode, each process has a dedicated ring
buffer, while in system mode, processes share a central buffer. To prevent
contention, how can we manage buffer assignment in system mode?
You are right, in system mode by default everything will go in the same
buffer.
You can declare more than one buffer, and repeat the data source assignment
by using the producer_name_filter. Something like:
data_sources {
target_buffer: 0
config {
name: "track_event"
}
producer_name_filter: "com.android.chrome"
producer_name_filter: "com.google.chrome.canary"
}
data_sources {
target_buffer: 1
config {
name: "track_event"
}
producer_name_filter: "another.process"
}
BTW, we sometimes see a CPU burst when we stop tracing, we definitely want
to store the trace, it will be nice if we could dump the trace with smooth
cpu usage :)
Unfortunately stopping the trace requires readback and a bunch of other
activities.
You could smooth it out over time if you use the write_into_file=true
setting, which writes the trace periodically on a file, rather than waiting
at the end.
But then you won't be able to have a ring buffer.
… —
Reply to this email directly, view it on GitHub
<#687>, or unsubscribe
<https://github.com/notifications/unsubscribe-auth/ABWORQIOOQM2DSBKAJJFLNLYPTSQRAVCNFSM6AAAAABCD3YLYWVHI2DSMVQWIX3LMV43ASLTON2WKOZSGA4TENJRGUYTEMA>
.
You are receiving this because you are subscribed to this thread.Message
ID: ***@***.***>
--
Primiano Tucci
Software Engineer
Google UK Limited
from perfetto.
dionysus1016
commented on May 31, 2024
@primiano Thanks for your answers. It's great to know that overhead is almost the same between inproc and system mode. Regarding the burst of CPU usage when dumping files, we plan to modify the code to work around it, for example, dumping entries in 50ms intervals.
I have another question about triggers. It seems the trigger is a session-level concept. For instance, when a trigger is activated, all data sources in the same session will apply the same rule to start/stop. Is it possible to configure different behavior per data source when the trigger is activated? For example, in one session, I want ftrace to stop without delay and track event stop after 1 second. To address this requirement, we plan to have a different session for different rules and then merge these trace files into a single one
from perfetto.
LalitMaganti
commented on May 31, 2024
Is it possible to configure different behavior per data source when the trigger is activated?
There's no easy way to do what you are asking about, triggers are a session wide concept and there are no plans to make them data-source specific.
While we generally do not advocate for merging traces, might be a reasonable option in your case.
from perfetto.
Related Issues (20)
- The recent updates seem to not support regular expression search. HOT 4
- Cannot write track events with custom datasource HOT 4
- tracing_muxer_impl.cc calls deprecated version of std::atomic_store HOT 8
- still racy prebuilt download HOT 11
- traced_probes crash HOT 4
- Question about the perfetto UI HOT 9
- How to turn off PERFETTO_WATCHDOG when compiling？ HOT 1
- Perfetto is crashed HOT 5
- Perfetto UI: Minimizing the size of deployed files? HOT 12
- Feature request: support serving Perfetto UI resources from a different directory HOT 7
- A wrong link in docs/reference /perfetto-cli.md
- Why does The perfetto UI display The Trace Processor instance on X.X.X. x:9002 is too old? HOT 7
- ToTimecode function is not work properly HOT 1
- How to quickly locate a ftrace slice on the timeline based on the start timestamp HOT 1
- Empty auxiliary tracks show up as crashed? HOT 2
- How to cut a large perfetto trace to some small file HOT 1
- Why can't I click the box corresponding to a single ftrace event? HOT 1
- Trace counters aren't present in timeline HOT 2
- traces with low timestamped logcat events fail to open properly HOT 1
- feature request: selectively display specific slice HOT 2
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from perfetto.