File bsc-1203931-17_drbd-only-reschedule-resync-quic... of Package drbd

Overview Repositories Revisions Requests Users Attributes Meta

File bsc-1203931-17_drbd-only-reschedule-resync-quickly-if-resync-sector.patch of Package drbd

From 330c8aa7b64f9fc0ace3c3985528b78aa0ca6780 Mon Sep 17 00:00:00 2001
From: Joel Colledge <joel.colledge@linbit.com>
Date: Thu, 4 Feb 2021 15:46:12 +0100
Subject: [PATCH] drbd: only reschedule resync quickly if resync sectors
 in-flight

8dc8ede32de44 drbd: fix slow sync when sync requests are answered quickly
introduced a mechanism to ensure that a resync does not become slow when
the responses are so fast that we have nothing to request. However, when
there are still in-flight sectors from a previous cycle, there is no
need to reschedule quickly. Instead we can just wait for the in-flight
sectors to arrive.

This situation can occur, for instance, when we have a large value for
c-fill-target and hit the "max-buffers in-flight" restriction. Without
this fix, the resync work was rescheduled repeatedly but made no new
requests until the responses to the older resync requests were received.
With this fix, DRBD does the next resync work after the standard delay
of 0.1s or when all resync responses are received.
---
 drbd/drbd_sender.c | 19 ++++++++++++++-----
 1 file changed, 14 insertions(+), 5 deletions(-)

diff --git a/drbd/drbd_sender.c b/drbd/drbd_sender.c
index aa1a3d6c1122..edf6b9d5c1f7 100644
--- a/drbd/drbd_sender.c
+++ b/drbd/drbd_sender.c
@@ -653,18 +653,28 @@ static int drbd_rs_number_requests(struct drbd_peer_device *peer_device)
 	return number;
 }
 
-static int drbd_single_request_delay(struct drbd_peer_device *peer_device)
+static int drbd_resync_delay(struct drbd_peer_device *peer_device)
 {
 	struct peer_device_conf *pdc;
 	unsigned long delay;
 
+	if (peer_device->rs_in_flight > 0) {
+		/* Requests in-flight. Use the standard delay. If all responses
+		 * are received before this time, the resync work will be
+		 * scheduled immediately. */
+		return RS_MAKE_REQS_INTV;
+	}
+
 	rcu_read_lock();
 	pdc = rcu_dereference(peer_device->conf);
 	if (rcu_dereference(peer_device->rs_plan_s)->size && pdc->c_max_rate > 0) {
-		/* The delay should be at least enough so that we can request
-		 * some data next time, so round up. */
+		/* Dynamic resync. This occurs when the peer responds so
+		 * quickly to the resync requests that the rate limiting
+		 * prevents any new requests from being made. Wait just long
+		 * enough so that we can request some data next time. */
 		delay = DIV_ROUND_UP(HZ * BM_SECT_PER_BIT, pdc->c_max_rate * 2);
 	} else {
+		/* Fixed resync rate. Use the standard delay. */
 		delay = RS_MAKE_REQS_INTV;
 	}
 	rcu_read_unlock();
@@ -862,8 +872,7 @@ request_done:
 			&peer_device->connection->sender_work,
 			&peer_device->resync_work);
 	} else {
-		unsigned long delay = number == 0 ? drbd_single_request_delay(peer_device) : RS_MAKE_REQS_INTV;
-		mod_timer(&peer_device->resync_timer, jiffies + delay);
+		mod_timer(&peer_device->resync_timer, jiffies + drbd_resync_delay(peer_device));
 	}
 	put_ldev(device);
 	return 0;
-- 
2.26.2

Places

File bsc-1203931-17_drbd-only-reschedule-resync-quickly-if-resync-sector.patch of Package drbd

Places