Quick info follow-up: strace-ing on the PIDs (with high CPU usage) shows a lot of EAGAIN errors from recvfrom() and a lot of recvfrom() + poll() + clock_gettime_monotonic() calls.
Thinking about it, would it make more sense to add a short wait instead [to reduce the throttle] and let it "infinitely" loop ? ["infinitely" loop == loop until it does not error with EAGAIN] On Tue, Jul 29, 2014 at 4:17 PM, Alexandru Ardelean <ardeleana...@gmail.com> wrote: > Seems that every once in a while libubus takes up the whole CPU > for certain processes. > > This could be bad code that I keep (re)writing. > > I don't know if it's ok to give up after a certain number of > retries; high CPU usage could indicate I should look deeper > at my logic. > > Which is why I've submitted this patch for review/opinions/comments. > > Can also be reviewed here: > https://github.com/commodo/ubus/commits/max_retry_mechanism > > Thanks > > --- > libubus-io.c | 22 ++++++++++++++++------ > 1 file changed, 16 insertions(+), 6 deletions(-) > > diff --git a/libubus-io.c b/libubus-io.c > index 31dad27..6bc06aa 100644 > --- a/libubus-io.c > +++ b/libubus-io.c > @@ -29,6 +29,7 @@ > #define STATIC_IOV(_var) { .iov_base = (char *) &(_var), .iov_len = > sizeof(_var) } > > #define UBUS_MSGBUF_REDUCTION_INTERVAL 16 > +#define UBUS_MAX_RETRIES 256 > > static const struct blob_attr_info ubus_policy[UBUS_ATTR_MAX] = { > [UBUS_ATTR_STATUS] = { .type = BLOB_ATTR_INT32 }, > @@ -75,6 +76,7 @@ static int writev_retry(int fd, struct iovec *iov, int > iov_len, int sock_fd) > .msg_controllen = sizeof(fd_buf), > }; > int len = 0; > + int retries = UBUS_MAX_RETRIES; > > do { > int cur_len; > @@ -91,9 +93,9 @@ static int writev_retry(int fd, struct iovec *iov, int > iov_len, int sock_fd) > switch(errno) { > case EAGAIN: > wait_data(fd, true); > - break; > case EINTR: > - break; > + if (retries-- > 0) > + break; > default: > return -1; > } > @@ -115,6 +117,7 @@ static int writev_retry(int fd, struct iovec *iov, int > iov_len, int sock_fd) > iov->iov_len -= cur_len; > msghdr.msg_iov = iov; > msghdr.msg_iovlen = iov_len; > + retries = UBUS_MAX_RETRIES; > } while (1); > > /* Should never reach here */ > @@ -170,6 +173,7 @@ static int recv_retry(int fd, struct iovec *iov, bool > wait, int *recv_fd) > .msg_iov = iov, > .msg_iovlen = 1, > }; > + int retries = UBUS_MAX_RETRIES; > > while (iov->iov_len > 0) { > if (wait) > @@ -192,11 +196,16 @@ static int recv_retry(int fd, struct iovec *iov, > bool wait, int *recv_fd) > bytes = 0; > if (uloop_cancelled) > return 0; > - if (errno == EINTR) > - continue; > - > - if (errno != EAGAIN) > + switch (errno) { > + case EINTR: > + if (retries-- > 0) > + continue; > + case EAGAIN: > + if (retries-- > 0) > + break; > + default: > return -1; > + } > } > if (!wait && !bytes) > return 0; > @@ -210,6 +219,7 @@ static int recv_retry(int fd, struct iovec *iov, bool > wait, int *recv_fd) > iov->iov_len -= bytes; > iov->iov_base += bytes; > total += bytes; > + retries = UBUS_MAX_RETRIES; > } > > return total; > -- > 1.8.4.5 > >
_______________________________________________ openwrt-devel mailing list openwrt-devel@lists.openwrt.org https://lists.openwrt.org/cgi-bin/mailman/listinfo/openwrt-devel
