summaryrefslogtreecommitdiffstats
path: root/xlators/cluster
diff options
context:
space:
mode:
authorPranith Kumar K <pkarampu@redhat.com>2015-09-08 16:23:36 +0530
committerPranith Kumar Karampuri <pkarampu@redhat.com>2015-10-29 04:52:49 -0700
commit6bbce9b1a48d5d50a2044b4518270e952331f159 (patch)
tree0c3cb1038b7b7b22a884e87897f1a1916f350cdc /xlators/cluster
parent73f8a582e365ef43b2454f263b5ca91a6de0475e (diff)
cluster/ec: Implement gfid-hash read-policy
Add a policy in ec to performs reads from same bricks as long as they are good. Based on the gfid of the file/directory it determines the bricks to be considered for reading. >Change-Id: Ic97b5c54c086a28b5e07a330a4fd448551b49376 >BUG: 1261260 >Signed-off-by: Pranith Kumar K <pkarampu@redhat.com> >Reviewed-on: http://review.gluster.org/12133 >Tested-by: NetBSD Build System <jenkins@build.gluster.org> >Tested-by: Gluster Build System <jenkins@build.gluster.com> >Reviewed-by: Xavier Hernandez <xhernandez@datalab.es> BUG: 1270705 Change-Id: Ibf0d21d7210125fa7aaa12b3f98bcdf7cd89ef02 Signed-off-by: Pranith Kumar K <pkarampu@redhat.com> Reviewed-on: http://review.gluster.org/12456 Tested-by: NetBSD Build System <jenkins@build.gluster.org> Tested-by: Gluster Build System <jenkins@build.gluster.com>
Diffstat (limited to 'xlators/cluster')
-rw-r--r--xlators/cluster/ec/src/ec-common.c37
-rw-r--r--xlators/cluster/ec/src/ec.c39
-rw-r--r--xlators/cluster/ec/src/ec.h7
3 files changed, 73 insertions, 10 deletions
diff --git a/xlators/cluster/ec/src/ec-common.c b/xlators/cluster/ec/src/ec-common.c
index 616b57232f3..7f1c3c535fa 100644
--- a/xlators/cluster/ec/src/ec-common.c
+++ b/xlators/cluster/ec/src/ec-common.c
@@ -9,6 +9,7 @@
*/
#include "byte-order.h"
+#include "hashfn.h"
#include "ec-mem-types.h"
#include "ec-data.h"
@@ -20,6 +21,25 @@
#include "ec.h"
#include "ec-messages.h"
+uint32_t
+ec_select_first_by_read_policy (ec_t *ec, ec_fop_data_t *fop)
+{
+ if (ec->read_policy == EC_ROUND_ROBIN) {
+ return ec->idx;
+ } else if (ec->read_policy == EC_GFID_HASH) {
+ if (fop->use_fd) {
+ return SuperFastHash((char *)fop->fd->inode->gfid,
+ sizeof(fop->fd->inode->gfid)) % ec->nodes;
+ } else {
+ if (gf_uuid_is_null (fop->loc[0].gfid))
+ loc_gfid (&fop->loc[0], fop->loc[0].gfid);
+ return SuperFastHash((char *)fop->loc[0].gfid,
+ sizeof(fop->loc[0].gfid)) % ec->nodes;
+ }
+ }
+ return 0;
+}
+
int32_t ec_child_valid(ec_t * ec, ec_fop_data_t * fop, int32_t idx)
{
return (idx < ec->nodes) && (((fop->remaining >> idx) & 1) == 1);
@@ -415,12 +435,13 @@ int32_t ec_child_select(ec_fop_data_t * fop)
fop->minimum = 1;
}
- first = ec->idx;
- if (++first >= ec->nodes)
- {
- first = 0;
+ if (ec->read_policy == EC_ROUND_ROBIN) {
+ first = ec->idx;
+ if (++first >= ec->nodes) {
+ first = 0;
+ }
+ ec->idx = first;
}
- ec->idx = first;
/*Unconditionally wind on healing subvolumes*/
fop->mask |= fop->healing;
@@ -518,14 +539,12 @@ void ec_dispatch_start(ec_fop_data_t * fop)
void ec_dispatch_one(ec_fop_data_t * fop)
{
- ec_t * ec = fop->xl->private;
-
ec_dispatch_start(fop);
if (ec_child_select(fop))
{
fop->expected = 1;
- fop->first = ec->idx;
+ fop->first = ec_select_first_by_read_policy (fop->xl->private, fop);
ec_dispatch_next(fop, fop->first);
}
@@ -589,7 +608,7 @@ void ec_dispatch_min(ec_fop_data_t * fop)
if (ec_child_select(fop))
{
fop->expected = count = ec->fragments;
- fop->first = ec->idx;
+ fop->first = ec_select_first_by_read_policy (fop->xl->private, fop);
idx = fop->first - 1;
mask = 0;
while (count-- > 0)
diff --git a/xlators/cluster/ec/src/ec.c b/xlators/cluster/ec/src/ec.c
index 7abd5cf8fa7..f139482c705 100644
--- a/xlators/cluster/ec/src/ec.c
+++ b/xlators/cluster/ec/src/ec.c
@@ -21,6 +21,11 @@
#include "ec-messages.h"
#include "ec-heald.h"
+static char *ec_read_policies[EC_READ_POLICY_MAX + 1] = {
+ [EC_ROUND_ROBIN] = "round-robin",
+ [EC_GFID_HASH] = "gfid-hash",
+ [EC_READ_POLICY_MAX] = NULL
+};
#define EC_MAX_FRAGMENTS EC_METHOD_MAX_FRAGMENTS
/* The maximum number of nodes is derived from the maximum allowed fragments
* using the rule that redundancy cannot be equal or greater than the number
@@ -231,10 +236,24 @@ ec_configure_background_heal_opts (ec_t *ec, int background_heals,
ec->background_heals = background_heals;
}
+int
+ec_assign_read_policy (ec_t *ec, char *read_policy)
+{
+ int read_policy_idx = -1;
+
+ read_policy_idx = gf_get_index_by_elem (ec_read_policies, read_policy);
+ if (read_policy_idx < 0 || read_policy_idx >= EC_READ_POLICY_MAX)
+ return -1;
+
+ ec->read_policy = read_policy_idx;
+ return 0;
+}
+
int32_t
reconfigure (xlator_t *this, dict_t *options)
{
ec_t *ec = this->private;
+ char *read_policy = NULL;
uint32_t heal_wait_qlen = 0;
uint32_t background_heals = 0;
@@ -248,6 +267,10 @@ reconfigure (xlator_t *this, dict_t *options)
uint32, failed);
ec_configure_background_heal_opts (ec, background_heals,
heal_wait_qlen);
+ GF_OPTION_RECONF ("read-policy", read_policy, options, str, failed);
+ if (ec_assign_read_policy (ec, read_policy))
+ goto failed;
+
return 0;
failed:
return -1;
@@ -512,7 +535,8 @@ notify (xlator_t *this, int32_t event, void *data, ...)
int32_t
init (xlator_t *this)
{
- ec_t *ec = NULL;
+ ec_t *ec = NULL;
+ char *read_policy = NULL;
if (this->parents == NULL)
{
@@ -574,6 +598,9 @@ init (xlator_t *this)
GF_OPTION_INIT ("heal-wait-qlength", ec->heal_wait_qlen, uint32, failed);
ec_configure_background_heal_opts (ec, ec->background_heals,
ec->heal_wait_qlen);
+ GF_OPTION_INIT ("read-policy", read_policy, str, failed);
+ if (ec_assign_read_policy (ec, read_policy))
+ goto failed;
if (ec->shd.iamshd)
ec_selfheal_daemon_init (this);
@@ -1189,6 +1216,7 @@ int32_t ec_dump_private(xlator_t *this)
gf_proc_dump_write("heal-wait-qlength", "%d", ec->heal_wait_qlen);
gf_proc_dump_write("healers", "%d", ec->healers);
gf_proc_dump_write("heal-waiters", "%d", ec->heal_waiters);
+ gf_proc_dump_write("read-policy", "%s", ec_read_policies[ec->read_policy]);
return 0;
}
@@ -1288,5 +1316,14 @@ struct volume_options options[] =
.description = "This option can be used to control number of heals"
" that can wait",
},
+ { .key = {"read-policy" },
+ .type = GF_OPTION_TYPE_STR,
+ .value = {"round-robin", "gfid-hash"},
+ .default_value = "round-robin",
+ .description = "inode-read fops happen only on 'k' number of bricks in"
+ " n=k+m disperse subvolume. 'round-robin' selects the read"
+ " subvolume using round-robin algo. 'gfid-hash' selects read"
+ " subvolume based on hash of the gfid of that file/directory.",
+ },
{ }
};
diff --git a/xlators/cluster/ec/src/ec.h b/xlators/cluster/ec/src/ec.h
index f335fd52afc..4ee7983b289 100644
--- a/xlators/cluster/ec/src/ec.h
+++ b/xlators/cluster/ec/src/ec.h
@@ -25,6 +25,12 @@
#define EC_VERSION_SIZE 2
+typedef enum {
+ EC_ROUND_ROBIN,
+ EC_GFID_HASH,
+ EC_READ_POLICY_MAX
+} ec_read_policy_t;
+
struct _ec
{
xlator_t * xl;
@@ -58,6 +64,7 @@ struct _ec
ec_self_heald_t shd;
char vol_uuid[UUID_SIZE + 1];
dict_t *leaf_to_subvolid;
+ ec_read_policy_t read_policy;
};
void ec_pending_fops_completed(ec_t *ec);