Re: [libvirt RFCv3] virfile: set pipe size in virFileWrapperFdNew to improve throughput

On 3/25/22 2:13 PM, Daniel P. Berrangé wrote: > On Fri, Mar 25, 2022 at 01:54:51PM +0100, Claudio Fontana wrote: >> currently the only user of virFileWrapperFdNew is the qemu driver; >> virsh save is very slow with a default pipe size. >> This change improves throughput by ~400% on fast nvme or ramdisk. >> >> Best value currently measured is 1MB, which happens to be also >> the kernel default for the pipe-max-size. >> >> Signed-off-by: Claudio Fontana <cfontana(a)suse.de&gt; >> --- >> >> see v2 at >> https://listman.redhat.com/archives/libvir-list/2022-March/229423.html >> >> Changes v2 -> v3: >> >> * removed reading of max-pipe-size from procfs, >> instead make multiple attempts on EPERM with smaller sizes. >> In the regular case, this should succeed on the first try. >> (Daniel) >> >> Changes v1 -> v2: >> >> * removed VIR_FILE_WRAPPER_BIG_PIPE, made the new pipe resizing >> unconditional (Michal) >> >> * moved code to separate functions (Michal) >> >> * removed ternary op, disliked in libvirt (Michal) >> >> * added #ifdef __linux__ (Ani Sinha) >> >> * try smallest value between currently best measured value (1MB) >> and the pipe-max-size setting. If pipe-max-size cannot be read, >> try kernel default max (1MB). (Daniel) >> >> >> src/util/virfile.c | 49 ++++++++++++++++++++++++++++++++++++++++++++++ >> 1 file changed, 49 insertions(+) >> >> diff --git a/src/util/virfile.c b/src/util/virfile.c >> index a04f888e06..876b865974 100644 >> --- a/src/util/virfile.c >> +++ b/src/util/virfile.c >> @@ -201,6 +201,51 @@ struct _virFileWrapperFd { >> }; >> >> #ifndef WIN32 >> + >> +#ifdef __linux__ >> + >> +/** >> + * virFileWrapperSetPipeSize: >> + * @fd: the fd of the pipe >> + * >> + * Set best pipe size on the passed file descriptor for bulk transfers of data. >> + * >> + * default pipe size (usually 64K) is generally not suited for large transfers >> + * to fast devices. A value of 1MB has been measured to improve virsh save >> + * by 400% in ideal conditions. We retry multiple times with smaller sizes >> + * on EPERM to account for possible small values of /proc/sys/fs/pipe-max-size. >> + * >> + * Return value is 0 on success, -1 and errno set on error. >> + * OS note: only for linux, on other OS this is a no-op. >> + */ >> +static int >> +virFileWrapperSetPipeSize(int fd) >> +{ >> + int sz; >> + >> + for (sz = 1024 * 1024; sz >= 64 * 1024; sz /= 2) { >> + int rv = fcntl(fd, F_SETPIPE_SZ, sz); >> + if (rv < 0 && errno == EPERM) { >> + continue; /* retry with half the size */ >> + } >> + if (rv < 0) { >> + break; >> + } >> + VIR_INFO("fd %d pipe size adjusted to %d", fd, sz); >> + return 0; >> + } >> + VIR_WARN("failed to set pipe size to %d (errno=%d)", sz, errno); >> + return -1; >> +} >> + >> +#else /* !__linux__ */ >> +static int virFileWrapperSetPipeSize(int fd) >> +{ >> + return 0; >> +} >> +#endif /* !__linux__ */ >> + >> + >> /** >> * virFileWrapperFdNew: >> * @fd: pointer to fd to wrap >> @@ -282,6 +327,10 @@ virFileWrapperFdNew(int *fd, const char *name, unsigned int flags) >> >> ret->cmd = virCommandNewArgList(iohelper_path, name, NULL); >> >> + if (virFileWrapperSetPipeSize(pipefd[!output]) < 0) { >> + virReportError(VIR_ERR_SYSTEM_ERROR, "%s", _("unable to set pipe size, data transfer might be slow")); > > Push this into virFileWrapperSetPipeSize instead of the VIR_WARN > there, and use virReportSystemError passing in the errno value too. ok, what about also warning on EPERM? In the normal case we should succeed on the first try I think.