On 11/21/19 1:24 PM, Xiubo Li wrote:
On 2019/11/21 10:46, Yan, Zheng wrote:
On 11/20/19 4:29 PM, xiubli@xxxxxxxxxx wrote:
From: Xiubo Li <xiubli@xxxxxxxxxx>
Even the MDS is in up:active state, but it also maybe laggy. Here
will skip the laggy MDSs.
Signed-off-by: Xiubo Li <xiubli@xxxxxxxxxx>
---
fs/ceph/mds_client.c | 6 ++++--
fs/ceph/mdsmap.c | 13 +++++++++----
2 files changed, 13 insertions(+), 6 deletions(-)
diff --git a/fs/ceph/mds_client.c b/fs/ceph/mds_client.c
index 82a929084671..a4e7026aaec9 100644
--- a/fs/ceph/mds_client.c
+++ b/fs/ceph/mds_client.c
@@ -972,7 +972,8 @@ static int __choose_mds(struct ceph_mds_client
*mdsc,
frag.frag, mds,
(int)r, frag.ndist);
if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
- CEPH_MDS_STATE_ACTIVE)
+ CEPH_MDS_STATE_ACTIVE &&
+ !ceph_mdsmap_is_laggy(mdsc->mdsmap, mds))
goto out;
}
@@ -987,7 +988,8 @@ static int __choose_mds(struct ceph_mds_client
*mdsc,
"frag %u mds%d (auth)\n",
inode, ceph_vinop(inode), frag.frag, mds);
if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
- CEPH_MDS_STATE_ACTIVE)
+ CEPH_MDS_STATE_ACTIVE &&
+ !ceph_mdsmap_is_laggy(mdsc->mdsmap, mds))
goto out;
}
}
for use USE_AUTH_MDS case, request can only be handled by auth mds.
client should send request to auth mds even it seems laggy.
BTW, what if the coreesponding auth mds was down, will it allow to
choose other mds ? From the current code it seems might. Or as long as
when the corresponding auth mds is in up:active state will the requests
only could to be handled by it ?
Some requests can only be handled by given MDS. Choosing other mds just
wastes resource.
Thanks.
diff --git a/fs/ceph/mdsmap.c b/fs/ceph/mdsmap.c
index 8b4f93e5b468..098669e6f1e4 100644
--- a/fs/ceph/mdsmap.c
+++ b/fs/ceph/mdsmap.c
@@ -13,6 +13,7 @@
#include "super.h"
+#define CEPH_MDS_IS_READY(i) (m->m_info[i].state > 0 &&
!m->m_info[i].laggy)
/*
* choose a random mds that is "up" (i.e. has a state > 0), or -1.
@@ -23,12 +24,16 @@ int ceph_mdsmap_get_random_mds(struct ceph_mdsmap
*m)
int i, j;
/* special case for one mds */
- if (1 == m->m_num_mds && m->m_info[0].state > 0)
- return 0;
+ if (1 == m->m_num_mds && m->m_info[0].state > 0) {
+ if (m->m_info[0].laggy)
+ return -1;
+ else
+ return 0;
+ }
/* count */
for (i = 0; i < m->m_num_mds; i++)
- if (m->m_info[i].state > 0)
+ if (CEPH_MDS_IS_READY(i))
n++;
if (n == 0)
return -1;
@@ -36,7 +41,7 @@ int ceph_mdsmap_get_random_mds(struct ceph_mdsmap *m)
/* pick */
n = prandom_u32() % n;
for (j = 0, i = 0; i < m->m_num_mds; i++) {
- if (m->m_info[i].state > 0)
+ if (CEPH_MDS_IS_READY(i))
j++;
if (j > n)
break;
