Return-Path: Received: from m9784.mail.qiye.163.com (m9784.mail.qiye.163.com [220.181.97.84]) by mail19.linbit.com (LINBIT Mail Daemon) with ESMTP id 8AD5A4207A7 for ; Mon, 9 Nov 2020 07:58:03 +0100 (CET) Received: from [192.168.9.105] (unknown [101.207.233.66]) by m9784.mail.qiye.163.com (Hmail) with ESMTPA id 1F4D041872 for ; Mon, 9 Nov 2020 14:52:08 +0800 (CST) From: Zhang Duan To: drbd-dev@lists.linbit.com Message-ID: <68d1f053-e293-c7c0-b974-cb47d95bdac5@easystack.cn> Date: Mon, 9 Nov 2020 14:52:07 +0800 MIME-Version: 1.0 Content-Type: multipart/alternative; boundary="------------DF3B984C51EA3A2A1BB945D5" Subject: [Drbd-dev] [PATCH] drbd: reply cancel to source when target of verify becomes inconsistent List-Id: "*Coordination* of development, patches, contributions -- *Questions* \(even to developers\) go to drbd-user, please." List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , This is a multi-part message in MIME format. --------------DF3B984C51EA3A2A1BB945D5 Content-Type: text/plain; charset=gbk; format=flowed Content-Transfer-Encoding: 7bit Consider one situation below: 1. primary A disconnects with secondary B & C both of which are Uptodate 2. start a verify from B to C 3. during the verify above, A reconnects and starts a resync to C because of primary-lost-quorum 4. C becomes resync target and Inconsistent, then ignore the verify requests from B which are already in its receive buffer 5. verify from B to C stalls here To resolve it, at step 4, C should reply a message of P_RS_CANCEL to B Here is the result if do that: Nov 6 16:25:46 node-2 kernel: drbd drbd1/0 drbd1 node-3: Skipped verify, too busy: start=170208, size=48 (sectors) Nov 6 16:25:46 node-2 kernel: drbd drbd1/0 drbd1 node-3: Online verify done but 6 4k blocks skipped (total 314 sec; paused 0 sec; 3336 K/sec) Nov 6 16:25:46 node-2 kernel: drbd drbd1/0 drbd1 node-3: repl( VerifyS -> Established ) Signed-off-by: ZhangDuan --- drbd/drbd_receiver.c | 2 ++ 1 file changed, 2 insertions(+) diff --git a/drbd/drbd_receiver.c b/drbd/drbd_receiver.c index 268dbf4d..c6d4b7c9 100644 --- a/drbd/drbd_receiver.c +++ b/drbd/drbd_receiver.c @@ -3273,6 +3273,8 @@ static int receive_DataRequest(struct drbd_connection *connection, struct packet break; case P_OV_REQUEST: verify_skipped_block(peer_device, sector, size); + drbd_send_ack_rp(peer_device, P_RS_CANCEL, p); + break; /* Fall through */ case P_RS_THIN_REQ: case P_RS_DATA_REQUEST: -- 2.24.0.windows.2 -- Sincerely Yours, Zhang Duan --------------DF3B984C51EA3A2A1BB945D5 Content-Type: text/html; charset=gbk Content-Transfer-Encoding: 7bit Consider one situation below:
1. primary A disconnects with secondary B & C both of which are Uptodate
2. start a verify from B to C
3. during the verify above, A reconnects and starts a resync to C because of
primary-lost-quorum
4. C becomes resync target and Inconsistent, then ignore the verify requests
from B which are already in its receive buffer
5. verify from B to C stalls here

To resolve it, at step 4, C should reply a message of P_RS_CANCEL to B
Here is the result if do that:
Nov 6 16:25:46 node-2 kernel: drbd drbd1/0 drbd1 node-3: Skipped verify, too busy: start=170208, size=48 (sectors)
Nov 6 16:25:46 node-2 kernel: drbd drbd1/0 drbd1 node-3: Online verify done but 6 4k blocks skipped (total 314 sec; paused 0 sec; 3336 K/sec)
Nov 6 16:25:46 node-2 kernel: drbd drbd1/0 drbd1 node-3: repl( VerifyS -> Established )

Signed-off-by: ZhangDuan
---
drbd/drbd_receiver.c | 2 ++
1 file changed, 2 insertions(+)

diff --git a/drbd/drbd_receiver.c b/drbd/drbd_receiver.c
index 268dbf4d..c6d4b7c9 100644
--- a/drbd/drbd_receiver.c
+++ b/drbd/drbd_receiver.c
@@ -3273,6 +3273,8 @@ static int receive_DataRequest(struct drbd_connection *connection, struct packet
break;
case P_OV_REQUEST:
verify_skipped_block(peer_device, sector, size);
+ drbd_send_ack_rp(peer_device, P_RS_CANCEL, p);
+ break;
/* Fall through */
case P_RS_THIN_REQ:
case P_RS_DATA_REQUEST:
--
2.24.0.windows.2

-- 
Sincerely Yours,
Zhang Duan
--------------DF3B984C51EA3A2A1BB945D5--