On Mon, Feb 07, 2011 at 11:18:02PM -0500, Martin Langhoff wrote: > On Mon, Feb 7, 2011 at 7:57 PM, James Cameron <[email protected]> wrote: > > Even with the buffered I/O, occasional writes will stall, resulting > > in activity code stall. > > In normal (!debug) mode logs are *tiny*. We have the right tradeoff.
No, I think you misunderstand. It doesn't matter how tiny the writes are, or how few there are, they still happen, and since they are to a filesystem, the kernel has the right to convert the write into an effectively synchronous operation. This happens when there is even trivial memory pressure. On XO-1.5, this hurts us badly, because of the random write performance of the internal storage. > > Future redesign might be to pipe the stdout/stderr to the Sugar shell > > and have it write it to filesystem only if the activity fails. ?;-) > > NO. That would be very bad. Activities can bring Sugar down (just > spotted one nasty that brings down X.org), and then you have no hint. If the writes were to a buffered pipe, as part of python's subprocess implementation, then the writes won't be delayed, since the buffer is already available. The text from the activity would not be processed, only written. I don't see how this could bring Sugar down, please provide more detail. A separate disposable logger process might alleviate your fear. It's something for Sugar Labs to consider. In the meanwhile we might log to tmpfs, with periodic rsync to storage. -- James Cameron http://quozl.linux.org.au/ _______________________________________________ Devel mailing list [email protected] http://lists.laptop.org/listinfo/devel
