On Wed, 21 Dec 2011, Brian Swetland wrote:
On Wed, Dec 21, 2011 at 8:47 PM, <david@xxxxxxx> wrote:
On Wed, 21 Dec 2011, Brian Swetland wrote:
Any thoughts as to how one could allow N userspace agents to log to a
single shared buffer without one agent, if buggy or malicious,
spamming out all the other contents of the log? This is one of the
main reasons we maintain separate logs in Android today, and are
likely to continue doing so until we figure out how resolve the issue.
note that you still don't prevent one app from blowing out the logs, all you
do is separate the logs into a handfulof categories and limit the app
(through standard filesystem permissions) to only blowing out one category
of logs.
or do I misunderstand this and you actually keep a separate queue for each
pid (which would seem to be a problem as you no longer know how much space
you need)
You're correct -- that's the model at present. We'd like to move to a
model where we have better control and are kicking around ideas for
how to do this for the next platform version.
The rate at which apps push data into logs is pretty amazing at times.
The system booting (maybe 10-20 services and 10+ apps starting up)
can blow through 256K of ringbuffer in seconds.
Having enough context available when something crashes to be able to
diagnose what went wrong, especially when problems could involve
interaction between multiple agents in multiple processes is important
for the frameworks and apps teams.
this makes it sound more like this belongs in userspace instead of the
kernel. if something crashes and they can flip a flag in userspace to say
'keep more logs now' and try to duplicate the crash is going to be far
easier than having to compile a new kernel to extend the size of the log.
Trying to get individual developers to be frugal with their logging
has turned out to be mostly a losing battle.
that's for sure, anything that's success depends on developers doing
things right that aren't visible to the user are pretty much doomed.
David Lang
