* [PATCH v3 0/2] COLO: handling of the full primary or secondary queue @ 2020-03-25 9:43 Derek Su 2020-03-25 9:43 ` [PATCH v3 1/2] net/colo-compare.c: Fix memory leak in packet_enqueue() Derek Su 2020-03-25 9:43 ` [PATCH v3 2/2] net/colo-compare.c: handling of the full primary or secondary queue Derek Su 0 siblings, 2 replies; 7+ messages in thread From: Derek Su @ 2020-03-25 9:43 UTC (permalink / raw) To: qemu-devel Cc: lizhijian, chyang, Derek Su, jasowang, ctcheng, chen.zhang, jwsu1986 The series is to handle the full primary or secondary queue in colo-compare. (1) fix the "pkt" memory leak in packet_enqueue(). Reproduce steps: 1. Setup PVM and SVM both with NIC e1000 by the steps descripted in the wiki qemu/COLO 2. Run "iperf3 -s" in PVM 3. Run "iperf3 -c <PVM-IP> -t 7200" in client The memory usage of qemu-system-x86_64 increases as the PVM's QMP shows "qemu-system-x86_64: colo compare secondary queue size too big, drop packet". (2) The pervious handling of the full primary or queue is only dropping the packet. If there are lots of clients to the guest VM, the "drop" will lead to the lost of the networking connection until next checkpoint, and therefore slow down the service. This patch drops the packet firstly. Then, send all queued primary packets, remove all queued secondary packets and do checkpoint. Please review, thanks. V3: - handling of the full primary or secondary queue according to the suggestion Zhang Chen V2: - Fix incorrect patch format Derek Su (2): net/colo-compare.c: Fix memory leak in packet_enqueue() net/colo-compare.c: handling of the full primary or secondary queue net/colo-compare.c | 64 ++++++++++++++++++++++++++++++++-------------- 1 file changed, 45 insertions(+), 19 deletions(-) -- 2.17.1 ^ permalink raw reply [flat|nested] 7+ messages in thread
* [PATCH v3 1/2] net/colo-compare.c: Fix memory leak in packet_enqueue() 2020-03-25 9:43 [PATCH v3 0/2] COLO: handling of the full primary or secondary queue Derek Su @ 2020-03-25 9:43 ` Derek Su 2020-03-25 9:43 ` [PATCH v3 2/2] net/colo-compare.c: handling of the full primary or secondary queue Derek Su 1 sibling, 0 replies; 7+ messages in thread From: Derek Su @ 2020-03-25 9:43 UTC (permalink / raw) To: qemu-devel Cc: lizhijian, chyang, Derek Su, jasowang, ctcheng, chen.zhang, jwsu1986 The patch is to fix the "pkt" memory leak in packet_enqueue(). The allocated "pkt" needs to be freed if the colo compare primary or secondary queue is too big. Signed-off-by: Derek Su <dereksu@qnap.com> --- net/colo-compare.c | 23 +++++++++++++++-------- 1 file changed, 15 insertions(+), 8 deletions(-) diff --git a/net/colo-compare.c b/net/colo-compare.c index 7ee17f2cf8..cdd87b2aa8 100644 --- a/net/colo-compare.c +++ b/net/colo-compare.c @@ -120,6 +120,10 @@ enum { SECONDARY_IN, }; +static const char *colo_mode[] = { + [PRIMARY_IN] = "primary", + [SECONDARY_IN] = "secondary", +}; static int compare_chr_send(CompareState *s, const uint8_t *buf, @@ -215,6 +219,7 @@ static int packet_enqueue(CompareState *s, int mode, Connection **con) ConnectionKey key; Packet *pkt = NULL; Connection *conn; + int ret; if (mode == PRIMARY_IN) { pkt = packet_new(s->pri_rs.buf, @@ -243,16 +248,18 @@ static int packet_enqueue(CompareState *s, int mode, Connection **con) } if (mode == PRIMARY_IN) { - if (!colo_insert_packet(&conn->primary_list, pkt, &conn->pack)) { - error_report("colo compare primary queue size too big," - "drop packet"); - } + ret = colo_insert_packet(&conn->primary_list, pkt, &conn->pack); } else { - if (!colo_insert_packet(&conn->secondary_list, pkt, &conn->sack)) { - error_report("colo compare secondary queue size too big," - "drop packet"); - } + ret = colo_insert_packet(&conn->secondary_list, pkt, &conn->sack); } + + if (!ret) { + error_report("colo compare %s queue size too big," + "drop packet", colo_mode[mode]); + packet_destroy(pkt, NULL); + pkt = NULL; + } + *con = conn; return 0; -- 2.17.1 ^ permalink raw reply related [flat|nested] 7+ messages in thread
* [PATCH v3 2/2] net/colo-compare.c: handling of the full primary or secondary queue 2020-03-25 9:43 [PATCH v3 0/2] COLO: handling of the full primary or secondary queue Derek Su 2020-03-25 9:43 ` [PATCH v3 1/2] net/colo-compare.c: Fix memory leak in packet_enqueue() Derek Su @ 2020-03-25 9:43 ` Derek Su 2020-03-27 17:45 ` Lukas Straub 1 sibling, 1 reply; 7+ messages in thread From: Derek Su @ 2020-03-25 9:43 UTC (permalink / raw) To: qemu-devel Cc: lizhijian, chyang, Derek Su, jasowang, ctcheng, chen.zhang, jwsu1986 The pervious handling of the full primary or queue is only dropping the packet. If there are lots of clients to the guest VM, the "drop" will lead to the lost of the networking connection until next checkpoint. To address the issue, this patch drops the packet firstly. Then, send all queued primary packets, remove all queued secondary packets and do checkpoint. Signed-off-by: Derek Su <dereksu@qnap.com> --- net/colo-compare.c | 41 ++++++++++++++++++++++++++++++----------- 1 file changed, 30 insertions(+), 11 deletions(-) diff --git a/net/colo-compare.c b/net/colo-compare.c index cdd87b2aa8..1a52f50fbe 100644 --- a/net/colo-compare.c +++ b/net/colo-compare.c @@ -125,6 +125,12 @@ static const char *colo_mode[] = { [SECONDARY_IN] = "secondary", }; +enum { + QUEUE_INSERT_ERR = -1, + QUEUE_INSERT_OK = 0, + QUEUE_INSERT_FULL = 1, +}; + static int compare_chr_send(CompareState *s, const uint8_t *buf, uint32_t size, @@ -211,8 +217,10 @@ static int colo_insert_packet(GQueue *queue, Packet *pkt, uint32_t *max_ack) } /* - * Return 0 on success, if return -1 means the pkt - * is unsupported(arp and ipv6) and will be sent later + * Return QUEUE_INSERT_OK on success. + * If return QUEUE_INSERT_FULL means list is full, and + * QUEUE_INSERT_ERR means the pkt is unsupported(arp and ipv6) and + * will be sent later */ static int packet_enqueue(CompareState *s, int mode, Connection **con) { @@ -234,7 +242,7 @@ static int packet_enqueue(CompareState *s, int mode, Connection **con) if (parse_packet_early(pkt)) { packet_destroy(pkt, NULL); pkt = NULL; - return -1; + return QUEUE_INSERT_ERR; } fill_connection_key(pkt, &key); @@ -258,11 +266,12 @@ static int packet_enqueue(CompareState *s, int mode, Connection **con) "drop packet", colo_mode[mode]); packet_destroy(pkt, NULL); pkt = NULL; + return QUEUE_INSERT_FULL; } *con = conn; - return 0; + return QUEUE_INSERT_OK; } static inline bool after(uint32_t seq1, uint32_t seq2) @@ -995,17 +1004,22 @@ static void compare_pri_rs_finalize(SocketReadState *pri_rs) { CompareState *s = container_of(pri_rs, CompareState, pri_rs); Connection *conn = NULL; + int ret; - if (packet_enqueue(s, PRIMARY_IN, &conn)) { + ret = packet_enqueue(s, PRIMARY_IN, &conn); + if (ret == QUEUE_INSERT_OK) { + /* compare packet in the specified connection */ + colo_compare_connection(conn, s); + } else if (ret == QUEUE_INSERT_FULL) { + g_queue_foreach(&s->conn_list, colo_flush_packets, s); + colo_compare_inconsistency_notify(s); + } else { trace_colo_compare_main("primary: unsupported packet in"); compare_chr_send(s, pri_rs->buf, pri_rs->packet_len, pri_rs->vnet_hdr_len, false); - } else { - /* compare packet in the specified connection */ - colo_compare_connection(conn, s); } } @@ -1013,12 +1027,17 @@ static void compare_sec_rs_finalize(SocketReadState *sec_rs) { CompareState *s = container_of(sec_rs, CompareState, sec_rs); Connection *conn = NULL; + int ret; - if (packet_enqueue(s, SECONDARY_IN, &conn)) { - trace_colo_compare_main("secondary: unsupported packet in"); - } else { + ret = packet_enqueue(s, SECONDARY_IN, &conn); + if (ret == QUEUE_INSERT_OK) { /* compare packet in the specified connection */ colo_compare_connection(conn, s); + } else if (ret == QUEUE_INSERT_FULL) { + g_queue_foreach(&s->conn_list, colo_flush_packets, s); + colo_compare_inconsistency_notify(s); + } else { + trace_colo_compare_main("secondary: unsupported packet in"); } } -- 2.17.1 ^ permalink raw reply related [flat|nested] 7+ messages in thread
* Re: [PATCH v3 2/2] net/colo-compare.c: handling of the full primary or secondary queue 2020-03-25 9:43 ` [PATCH v3 2/2] net/colo-compare.c: handling of the full primary or secondary queue Derek Su @ 2020-03-27 17:45 ` Lukas Straub 2020-03-27 18:20 ` Derek Su 0 siblings, 1 reply; 7+ messages in thread From: Lukas Straub @ 2020-03-27 17:45 UTC (permalink / raw) To: Derek Su Cc: lizhijian, chyang, jasowang, qemu-devel, ctcheng, chen.zhang, jwsu1986 [-- Attachment #1: Type: text/plain, Size: 4466 bytes --] On Wed, 25 Mar 2020 17:43:54 +0800 Derek Su <dereksu@qnap.com> wrote: > The pervious handling of the full primary or queue is only dropping > the packet. If there are lots of clients to the guest VM, > the "drop" will lead to the lost of the networking connection > until next checkpoint. > > To address the issue, this patch drops the packet firstly. > Then, send all queued primary packets, remove all queued secondary > packets and do checkpoint. > > Signed-off-by: Derek Su <dereksu@qnap.com> > --- > net/colo-compare.c | 41 ++++++++++++++++++++++++++++++----------- > 1 file changed, 30 insertions(+), 11 deletions(-) > > diff --git a/net/colo-compare.c b/net/colo-compare.c > index cdd87b2aa8..1a52f50fbe 100644 > --- a/net/colo-compare.c > +++ b/net/colo-compare.c > @@ -125,6 +125,12 @@ static const char *colo_mode[] = { > [SECONDARY_IN] = "secondary", > }; > > +enum { > + QUEUE_INSERT_ERR = -1, > + QUEUE_INSERT_OK = 0, > + QUEUE_INSERT_FULL = 1, > +}; > + > static int compare_chr_send(CompareState *s, > const uint8_t *buf, > uint32_t size, > @@ -211,8 +217,10 @@ static int colo_insert_packet(GQueue *queue, Packet *pkt, uint32_t *max_ack) > } > > /* > - * Return 0 on success, if return -1 means the pkt > - * is unsupported(arp and ipv6) and will be sent later > + * Return QUEUE_INSERT_OK on success. > + * If return QUEUE_INSERT_FULL means list is full, and > + * QUEUE_INSERT_ERR means the pkt is unsupported(arp and ipv6) and > + * will be sent later > */ > static int packet_enqueue(CompareState *s, int mode, Connection **con) > { > @@ -234,7 +242,7 @@ static int packet_enqueue(CompareState *s, int mode, Connection **con) > if (parse_packet_early(pkt)) { > packet_destroy(pkt, NULL); > pkt = NULL; > - return -1; > + return QUEUE_INSERT_ERR; > } > fill_connection_key(pkt, &key); > > @@ -258,11 +266,12 @@ static int packet_enqueue(CompareState *s, int mode, Connection **con) > "drop packet", colo_mode[mode]); > packet_destroy(pkt, NULL); > pkt = NULL; > + return QUEUE_INSERT_FULL; > } > > *con = conn; > > - return 0; > + return QUEUE_INSERT_OK; > } > > static inline bool after(uint32_t seq1, uint32_t seq2) > @@ -995,17 +1004,22 @@ static void compare_pri_rs_finalize(SocketReadState *pri_rs) > { > CompareState *s = container_of(pri_rs, CompareState, pri_rs); > Connection *conn = NULL; > + int ret; > > - if (packet_enqueue(s, PRIMARY_IN, &conn)) { > + ret = packet_enqueue(s, PRIMARY_IN, &conn); > + if (ret == QUEUE_INSERT_OK) { > + /* compare packet in the specified connection */ > + colo_compare_connection(conn, s); > + } else if (ret == QUEUE_INSERT_FULL) { > + g_queue_foreach(&s->conn_list, colo_flush_packets, s); > + colo_compare_inconsistency_notify(s); > + } else { > trace_colo_compare_main("primary: unsupported packet in"); > compare_chr_send(s, > pri_rs->buf, > pri_rs->packet_len, > pri_rs->vnet_hdr_len, > false); > - } else { > - /* compare packet in the specified connection */ > - colo_compare_connection(conn, s); > } > } > > @@ -1013,12 +1027,17 @@ static void compare_sec_rs_finalize(SocketReadState *sec_rs) > { > CompareState *s = container_of(sec_rs, CompareState, sec_rs); > Connection *conn = NULL; > + int ret; > > - if (packet_enqueue(s, SECONDARY_IN, &conn)) { > - trace_colo_compare_main("secondary: unsupported packet in"); > - } else { > + ret = packet_enqueue(s, SECONDARY_IN, &conn); > + if (ret == QUEUE_INSERT_OK) { > /* compare packet in the specified connection */ > colo_compare_connection(conn, s); > + } else if (ret == QUEUE_INSERT_FULL) { > + g_queue_foreach(&s->conn_list, colo_flush_packets, s); > + colo_compare_inconsistency_notify(s); > + } else { > + trace_colo_compare_main("secondary: unsupported packet in"); > } > } > Hi, I don't think we have to flush here because the (post-)checkpoint event will flush the packets for us. Regards, Lukas Straub [-- Attachment #2: OpenPGP digital signature --] [-- Type: application/pgp-signature, Size: 833 bytes --] ^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [PATCH v3 2/2] net/colo-compare.c: handling of the full primary or secondary queue 2020-03-27 17:45 ` Lukas Straub @ 2020-03-27 18:20 ` Derek Su 2020-03-27 18:28 ` Lukas Straub 0 siblings, 1 reply; 7+ messages in thread From: Derek Su @ 2020-03-27 18:20 UTC (permalink / raw) To: Lukas Straub Cc: lizhijian, chyang, Derek Su, jasowang, qemu-devel, ctcheng, Zhang, Chen Lukas Straub <lukasstraub2@web.de> 於 2020年3月28日 週六 上午1:46寫道: > > On Wed, 25 Mar 2020 17:43:54 +0800 > Derek Su <dereksu@qnap.com> wrote: > > > The pervious handling of the full primary or queue is only dropping > > the packet. If there are lots of clients to the guest VM, > > the "drop" will lead to the lost of the networking connection > > until next checkpoint. > > > > To address the issue, this patch drops the packet firstly. > > Then, send all queued primary packets, remove all queued secondary > > packets and do checkpoint. > > > > Signed-off-by: Derek Su <dereksu@qnap.com> > > --- > > net/colo-compare.c | 41 ++++++++++++++++++++++++++++++----------- > > 1 file changed, 30 insertions(+), 11 deletions(-) > > > > diff --git a/net/colo-compare.c b/net/colo-compare.c > > index cdd87b2aa8..1a52f50fbe 100644 > > --- a/net/colo-compare.c > > +++ b/net/colo-compare.c > > @@ -125,6 +125,12 @@ static const char *colo_mode[] = { > > [SECONDARY_IN] = "secondary", > > }; > > > > +enum { > > + QUEUE_INSERT_ERR = -1, > > + QUEUE_INSERT_OK = 0, > > + QUEUE_INSERT_FULL = 1, > > +}; > > + > > static int compare_chr_send(CompareState *s, > > const uint8_t *buf, > > uint32_t size, > > @@ -211,8 +217,10 @@ static int colo_insert_packet(GQueue *queue, Packet *pkt, uint32_t *max_ack) > > } > > > > /* > > - * Return 0 on success, if return -1 means the pkt > > - * is unsupported(arp and ipv6) and will be sent later > > + * Return QUEUE_INSERT_OK on success. > > + * If return QUEUE_INSERT_FULL means list is full, and > > + * QUEUE_INSERT_ERR means the pkt is unsupported(arp and ipv6) and > > + * will be sent later > > */ > > static int packet_enqueue(CompareState *s, int mode, Connection **con) > > { > > @@ -234,7 +242,7 @@ static int packet_enqueue(CompareState *s, int mode, Connection **con) > > if (parse_packet_early(pkt)) { > > packet_destroy(pkt, NULL); > > pkt = NULL; > > - return -1; > > + return QUEUE_INSERT_ERR; > > } > > fill_connection_key(pkt, &key); > > > > @@ -258,11 +266,12 @@ static int packet_enqueue(CompareState *s, int mode, Connection **con) > > "drop packet", colo_mode[mode]); > > packet_destroy(pkt, NULL); > > pkt = NULL; > > + return QUEUE_INSERT_FULL; > > } > > > > *con = conn; > > > > - return 0; > > + return QUEUE_INSERT_OK; > > } > > > > static inline bool after(uint32_t seq1, uint32_t seq2) > > @@ -995,17 +1004,22 @@ static void compare_pri_rs_finalize(SocketReadState *pri_rs) > > { > > CompareState *s = container_of(pri_rs, CompareState, pri_rs); > > Connection *conn = NULL; > > + int ret; > > > > - if (packet_enqueue(s, PRIMARY_IN, &conn)) { > > + ret = packet_enqueue(s, PRIMARY_IN, &conn); > > + if (ret == QUEUE_INSERT_OK) { > > + /* compare packet in the specified connection */ > > + colo_compare_connection(conn, s); > > + } else if (ret == QUEUE_INSERT_FULL) { > > + g_queue_foreach(&s->conn_list, colo_flush_packets, s); > > + colo_compare_inconsistency_notify(s); > > + } else { > > trace_colo_compare_main("primary: unsupported packet in"); > > compare_chr_send(s, > > pri_rs->buf, > > pri_rs->packet_len, > > pri_rs->vnet_hdr_len, > > false); > > - } else { > > - /* compare packet in the specified connection */ > > - colo_compare_connection(conn, s); > > } > > } > > > > @@ -1013,12 +1027,17 @@ static void compare_sec_rs_finalize(SocketReadState *sec_rs) > > { > > CompareState *s = container_of(sec_rs, CompareState, sec_rs); > > Connection *conn = NULL; > > + int ret; > > > > - if (packet_enqueue(s, SECONDARY_IN, &conn)) { > > - trace_colo_compare_main("secondary: unsupported packet in"); > > - } else { > > + ret = packet_enqueue(s, SECONDARY_IN, &conn); > > + if (ret == QUEUE_INSERT_OK) { > > /* compare packet in the specified connection */ > > colo_compare_connection(conn, s); > > + } else if (ret == QUEUE_INSERT_FULL) { > > + g_queue_foreach(&s->conn_list, colo_flush_packets, s); > > + colo_compare_inconsistency_notify(s); > > + } else { > > + trace_colo_compare_main("secondary: unsupported packet in"); > > } > > } > > > > Hi, > I don't think we have to flush here because the (post-)checkpoint event will flush the packets for us. > Hi, Yes, the periodical checkpoint can flush the packets. But, if many clients send many packets to the vm, there is a high probability that packets are dropped because the primary/secondary queues are always full. It causes lots of re-transmission between clients and vm and deteriorate service response to clients. Sincerely, Derek Su > Regards, > Lukas Straub ^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [PATCH v3 2/2] net/colo-compare.c: handling of the full primary or secondary queue 2020-03-27 18:20 ` Derek Su @ 2020-03-27 18:28 ` Lukas Straub 2020-03-28 3:35 ` Derek Su 0 siblings, 1 reply; 7+ messages in thread From: Lukas Straub @ 2020-03-27 18:28 UTC (permalink / raw) To: Derek Su Cc: lizhijian, chyang, Derek Su, jasowang, qemu-devel, ctcheng, Zhang, Chen [-- Attachment #1: Type: text/plain, Size: 5603 bytes --] On Sat, 28 Mar 2020 02:20:21 +0800 Derek Su <jwsu1986@gmail.com> wrote: > Lukas Straub <lukasstraub2@web.de> 於 2020年3月28日 週六 上午1:46寫道: > > > > On Wed, 25 Mar 2020 17:43:54 +0800 > > Derek Su <dereksu@qnap.com> wrote: > > > > > The pervious handling of the full primary or queue is only dropping > > > the packet. If there are lots of clients to the guest VM, > > > the "drop" will lead to the lost of the networking connection > > > until next checkpoint. > > > > > > To address the issue, this patch drops the packet firstly. > > > Then, send all queued primary packets, remove all queued secondary > > > packets and do checkpoint. > > > > > > Signed-off-by: Derek Su <dereksu@qnap.com> > > > --- > > > net/colo-compare.c | 41 ++++++++++++++++++++++++++++++----------- > > > 1 file changed, 30 insertions(+), 11 deletions(-) > > > > > > diff --git a/net/colo-compare.c b/net/colo-compare.c > > > index cdd87b2aa8..1a52f50fbe 100644 > > > --- a/net/colo-compare.c > > > +++ b/net/colo-compare.c > > > @@ -125,6 +125,12 @@ static const char *colo_mode[] = { > > > [SECONDARY_IN] = "secondary", > > > }; > > > > > > +enum { > > > + QUEUE_INSERT_ERR = -1, > > > + QUEUE_INSERT_OK = 0, > > > + QUEUE_INSERT_FULL = 1, > > > +}; > > > + > > > static int compare_chr_send(CompareState *s, > > > const uint8_t *buf, > > > uint32_t size, > > > @@ -211,8 +217,10 @@ static int colo_insert_packet(GQueue *queue, Packet *pkt, uint32_t *max_ack) > > > } > > > > > > /* > > > - * Return 0 on success, if return -1 means the pkt > > > - * is unsupported(arp and ipv6) and will be sent later > > > + * Return QUEUE_INSERT_OK on success. > > > + * If return QUEUE_INSERT_FULL means list is full, and > > > + * QUEUE_INSERT_ERR means the pkt is unsupported(arp and ipv6) and > > > + * will be sent later > > > */ > > > static int packet_enqueue(CompareState *s, int mode, Connection **con) > > > { > > > @@ -234,7 +242,7 @@ static int packet_enqueue(CompareState *s, int mode, Connection **con) > > > if (parse_packet_early(pkt)) { > > > packet_destroy(pkt, NULL); > > > pkt = NULL; > > > - return -1; > > > + return QUEUE_INSERT_ERR; > > > } > > > fill_connection_key(pkt, &key); > > > > > > @@ -258,11 +266,12 @@ static int packet_enqueue(CompareState *s, int mode, Connection **con) > > > "drop packet", colo_mode[mode]); > > > packet_destroy(pkt, NULL); > > > pkt = NULL; > > > + return QUEUE_INSERT_FULL; > > > } > > > > > > *con = conn; > > > > > > - return 0; > > > + return QUEUE_INSERT_OK; > > > } > > > > > > static inline bool after(uint32_t seq1, uint32_t seq2) > > > @@ -995,17 +1004,22 @@ static void compare_pri_rs_finalize(SocketReadState *pri_rs) > > > { > > > CompareState *s = container_of(pri_rs, CompareState, pri_rs); > > > Connection *conn = NULL; > > > + int ret; > > > > > > - if (packet_enqueue(s, PRIMARY_IN, &conn)) { > > > + ret = packet_enqueue(s, PRIMARY_IN, &conn); > > > + if (ret == QUEUE_INSERT_OK) { > > > + /* compare packet in the specified connection */ > > > + colo_compare_connection(conn, s); > > > + } else if (ret == QUEUE_INSERT_FULL) { > > > + g_queue_foreach(&s->conn_list, colo_flush_packets, s); > > > + colo_compare_inconsistency_notify(s); > > > + } else { > > > trace_colo_compare_main("primary: unsupported packet in"); > > > compare_chr_send(s, > > > pri_rs->buf, > > > pri_rs->packet_len, > > > pri_rs->vnet_hdr_len, > > > false); > > > - } else { > > > - /* compare packet in the specified connection */ > > > - colo_compare_connection(conn, s); > > > } > > > } > > > > > > @@ -1013,12 +1027,17 @@ static void compare_sec_rs_finalize(SocketReadState *sec_rs) > > > { > > > CompareState *s = container_of(sec_rs, CompareState, sec_rs); > > > Connection *conn = NULL; > > > + int ret; > > > > > > - if (packet_enqueue(s, SECONDARY_IN, &conn)) { > > > - trace_colo_compare_main("secondary: unsupported packet in"); > > > - } else { > > > + ret = packet_enqueue(s, SECONDARY_IN, &conn); > > > + if (ret == QUEUE_INSERT_OK) { > > > /* compare packet in the specified connection */ > > > colo_compare_connection(conn, s); > > > + } else if (ret == QUEUE_INSERT_FULL) { > > > + g_queue_foreach(&s->conn_list, colo_flush_packets, s); > > > + colo_compare_inconsistency_notify(s); > > > + } else { > > > + trace_colo_compare_main("secondary: unsupported packet in"); > > > } > > > } > > > > > > > Hi, > > I don't think we have to flush here because the (post-)checkpoint event will flush the packets for us. > > > > Hi, > Yes, the periodical checkpoint can flush the packets. > > But, if many clients send many packets to the vm, > there is a high probability that packets are dropped because the > primary/secondary queues are always full. > It causes lots of re-transmission between clients and vm and > deteriorate service response to clients. > > Sincerely, > Derek Su I mean that we can still initiate a checkpoint here, but not do the flushing here. Regards, Lukas Straub > > Regards, > > Lukas Straub [-- Attachment #2: OpenPGP digital signature --] [-- Type: application/pgp-signature, Size: 833 bytes --] ^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [PATCH v3 2/2] net/colo-compare.c: handling of the full primary or secondary queue 2020-03-27 18:28 ` Lukas Straub @ 2020-03-28 3:35 ` Derek Su 0 siblings, 0 replies; 7+ messages in thread From: Derek Su @ 2020-03-28 3:35 UTC (permalink / raw) To: Lukas Straub Cc: lizhijian, chyang, Derek Su, jasowang, qemu-devel, ctcheng, Zhang, Chen Lukas Straub <lukasstraub2@web.de> 於 2020年3月28日 週六 上午2:28寫道: > > On Sat, 28 Mar 2020 02:20:21 +0800 > Derek Su <jwsu1986@gmail.com> wrote: > > > Lukas Straub <lukasstraub2@web.de> 於 2020年3月28日 週六 上午1:46寫道: > > > > > > On Wed, 25 Mar 2020 17:43:54 +0800 > > > Derek Su <dereksu@qnap.com> wrote: > > > > > > > The pervious handling of the full primary or queue is only dropping > > > > the packet. If there are lots of clients to the guest VM, > > > > the "drop" will lead to the lost of the networking connection > > > > until next checkpoint. > > > > > > > > To address the issue, this patch drops the packet firstly. > > > > Then, send all queued primary packets, remove all queued secondary > > > > packets and do checkpoint. > > > > > > > > Signed-off-by: Derek Su <dereksu@qnap.com> > > > > --- > > > > net/colo-compare.c | 41 ++++++++++++++++++++++++++++++----------- > > > > 1 file changed, 30 insertions(+), 11 deletions(-) > > > > > > > > diff --git a/net/colo-compare.c b/net/colo-compare.c > > > > index cdd87b2aa8..1a52f50fbe 100644 > > > > --- a/net/colo-compare.c > > > > +++ b/net/colo-compare.c > > > > @@ -125,6 +125,12 @@ static const char *colo_mode[] = { > > > > [SECONDARY_IN] = "secondary", > > > > }; > > > > > > > > +enum { > > > > + QUEUE_INSERT_ERR = -1, > > > > + QUEUE_INSERT_OK = 0, > > > > + QUEUE_INSERT_FULL = 1, > > > > +}; > > > > + > > > > static int compare_chr_send(CompareState *s, > > > > const uint8_t *buf, > > > > uint32_t size, > > > > @@ -211,8 +217,10 @@ static int colo_insert_packet(GQueue *queue, Packet *pkt, uint32_t *max_ack) > > > > } > > > > > > > > /* > > > > - * Return 0 on success, if return -1 means the pkt > > > > - * is unsupported(arp and ipv6) and will be sent later > > > > + * Return QUEUE_INSERT_OK on success. > > > > + * If return QUEUE_INSERT_FULL means list is full, and > > > > + * QUEUE_INSERT_ERR means the pkt is unsupported(arp and ipv6) and > > > > + * will be sent later > > > > */ > > > > static int packet_enqueue(CompareState *s, int mode, Connection **con) > > > > { > > > > @@ -234,7 +242,7 @@ static int packet_enqueue(CompareState *s, int mode, Connection **con) > > > > if (parse_packet_early(pkt)) { > > > > packet_destroy(pkt, NULL); > > > > pkt = NULL; > > > > - return -1; > > > > + return QUEUE_INSERT_ERR; > > > > } > > > > fill_connection_key(pkt, &key); > > > > > > > > @@ -258,11 +266,12 @@ static int packet_enqueue(CompareState *s, int mode, Connection **con) > > > > "drop packet", colo_mode[mode]); > > > > packet_destroy(pkt, NULL); > > > > pkt = NULL; > > > > + return QUEUE_INSERT_FULL; > > > > } > > > > > > > > *con = conn; > > > > > > > > - return 0; > > > > + return QUEUE_INSERT_OK; > > > > } > > > > > > > > static inline bool after(uint32_t seq1, uint32_t seq2) > > > > @@ -995,17 +1004,22 @@ static void compare_pri_rs_finalize(SocketReadState *pri_rs) > > > > { > > > > CompareState *s = container_of(pri_rs, CompareState, pri_rs); > > > > Connection *conn = NULL; > > > > + int ret; > > > > > > > > - if (packet_enqueue(s, PRIMARY_IN, &conn)) { > > > > + ret = packet_enqueue(s, PRIMARY_IN, &conn); > > > > + if (ret == QUEUE_INSERT_OK) { > > > > + /* compare packet in the specified connection */ > > > > + colo_compare_connection(conn, s); > > > > + } else if (ret == QUEUE_INSERT_FULL) { > > > > + g_queue_foreach(&s->conn_list, colo_flush_packets, s); > > > > + colo_compare_inconsistency_notify(s); > > > > + } else { > > > > trace_colo_compare_main("primary: unsupported packet in"); > > > > compare_chr_send(s, > > > > pri_rs->buf, > > > > pri_rs->packet_len, > > > > pri_rs->vnet_hdr_len, > > > > false); > > > > - } else { > > > > - /* compare packet in the specified connection */ > > > > - colo_compare_connection(conn, s); > > > > } > > > > } > > > > > > > > @@ -1013,12 +1027,17 @@ static void compare_sec_rs_finalize(SocketReadState *sec_rs) > > > > { > > > > CompareState *s = container_of(sec_rs, CompareState, sec_rs); > > > > Connection *conn = NULL; > > > > + int ret; > > > > > > > > - if (packet_enqueue(s, SECONDARY_IN, &conn)) { > > > > - trace_colo_compare_main("secondary: unsupported packet in"); > > > > - } else { > > > > + ret = packet_enqueue(s, SECONDARY_IN, &conn); > > > > + if (ret == QUEUE_INSERT_OK) { > > > > /* compare packet in the specified connection */ > > > > colo_compare_connection(conn, s); > > > > + } else if (ret == QUEUE_INSERT_FULL) { > > > > + g_queue_foreach(&s->conn_list, colo_flush_packets, s); > > > > + colo_compare_inconsistency_notify(s); > > > > + } else { > > > > + trace_colo_compare_main("secondary: unsupported packet in"); > > > > } > > > > } > > > > > > > > > > Hi, > > > I don't think we have to flush here because the (post-)checkpoint event will flush the packets for us. > > > > > > > Hi, > > Yes, the periodical checkpoint can flush the packets. > > > > But, if many clients send many packets to the vm, > > there is a high probability that packets are dropped because the > > primary/secondary queues are always full. > > It causes lots of re-transmission between clients and vm and > > deteriorate service response to clients. > > > > Sincerely, > > Derek Su > > I mean that we can still initiate a checkpoint here, but not do the flushing here. > > Regards, > Lukas Straub Hi, Your're right. After checking the colo_do_checkpoint_transaction function, the flush in the patch is redundant. It will be fixed in the next version. Thanks. Derek Su > > > > Regards, > > > Lukas Straub > ^ permalink raw reply [flat|nested] 7+ messages in thread
end of thread, other threads:[~2020-03-28 3:36 UTC | newest] Thread overview: 7+ messages (download: mbox.gz follow: Atom feed -- links below jump to the message on this page -- 2020-03-25 9:43 [PATCH v3 0/2] COLO: handling of the full primary or secondary queue Derek Su 2020-03-25 9:43 ` [PATCH v3 1/2] net/colo-compare.c: Fix memory leak in packet_enqueue() Derek Su 2020-03-25 9:43 ` [PATCH v3 2/2] net/colo-compare.c: handling of the full primary or secondary queue Derek Su 2020-03-27 17:45 ` Lukas Straub 2020-03-27 18:20 ` Derek Su 2020-03-27 18:28 ` Lukas Straub 2020-03-28 3:35 ` Derek Su
This is a public inbox, see mirroring instructions for how to clone and mirror all data and code used for this inbox; as well as URLs for NNTP newsgroup(s).