Glib's internal buffering implementation is opaque. It seems to grow up
when receiving large chunk of data (eg. by guest-file-write). When that
buffer become large, the glib main loop will stop notifying callback for
small incoming chunk (eg. by guest-file-close) but wait until buffer is
filled up again. In this situation, qemu-ga will simply looks like
hang/dead from the client side of view.
By disabling the buffer, qemu-ga will deal with each incoming chunk with
no delay. So the client side will no longer face the situation where an
issued command doesn't come back with a response.
Signed-off-by: WANG Chao <wcw...@gmail.com>
---
qga/channel-posix.c | 9 +--------
1 file changed, 1 insertion(+), 8 deletions(-)
diff --git a/qga/channel-posix.c b/qga/channel-posix.c
index 8aad4fe..a2ca161 100644
--- a/qga/channel-posix.c
+++ b/qga/channel-posix.c
@@ -116,6 +116,7 @@ static int ga_channel_client_add(GAChannel *c, int fd)
client_channel = g_io_channel_unix_new(fd);
g_assert(client_channel);
g_io_channel_set_encoding(client_channel, NULL, &err);
+ g_io_channel_set_buffered(client_channel, false);
if (err != NULL) {
g_warning("error setting channel encoding to binary");
g_error_free(err);
@@ -230,14 +231,6 @@ GIOStatus ga_channel_write_all(GAChannel *c, const gchar
*buf, gsize size)
size -= written;
}
- if (status == G_IO_STATUS_NORMAL) {
- status = g_io_channel_flush(c->client_channel, &err);
- if (err != NULL) {
- g_warning("error flushing channel: %s", err->message);
- return G_IO_STATUS_ERROR;
- }
- }
-
return status;
}
--
2.3.0
