mirror/mariadb
1
0
Fork 0
mirror of https://github.com/MariaDB/server.git synced 2025-02-02 03:51:50 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions

[t:3703], merge to main

Browse source 
git-svn-id: file:///svn/toku/tokudb@32737 c7de825b-a66e-492c-adef-691d508d4ae1
This commit is contained in:
Zardosht Kasheff 2013-04-16 23:59:43 -04:00 • committed by Yoni Fogel
parent 4790f1e276
commit d5505938e9
3 changed files with 39 additions and 41 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

3
newbrt/brt-internal.h
View file

@ -140,7 +140,6 @@ struct brtnode_leaf_basement_node {
OMT buffer;
unsigned int n_bytes_in_buffer; /* How many bytes to represent the OMT (including the per-key overheads, but not including the overheads for the node. */
unsigned int seqinsert; /* number of sequential inserts to this leaf */
MSN max_msn_applied;
};
#define PT_INVALID 0
@ -301,7 +300,6 @@ static inline void set_BSB(BRTNODE node, int i, SUB_BLOCK sb) {
#define BNC_BUFFER(node,i) (BNC(node,i)->buffer)
#define BNC_NBYTESINBUF(node,i) (BNC(node,i)->n_bytes_in_buffer)
#define BNC_NBYTESINBUF(node,i) (BNC(node,i)->n_bytes_in_buffer)
// leaf node macros
#define BLB_OPTIMIZEDFORUPGRADE(node,i) (BLB(node,i)->optimized_for_upgrade)
@ -614,7 +612,6 @@ brt_leaf_apply_cmd_once (
TOKULOGGER logger,
uint64_t *workdonep
);
void brt_leaf_put_cmd (BRT t, BASEMENTNODE bn, SUBTREE_EST se, BRT_MSG cmd, bool *made_change, uint64_t *workdonep);
void
toku_apply_cmd_to_leaf(BRT t, BRTNODE node, BRT_MSG cmd, bool *made_change, uint64_t *workdonep);

2
newbrt/brt-serialize.c
View file

@ -804,7 +804,6 @@ BASEMENTNODE toku_create_empty_bn_no_buffer(void) {
bn->n_bytes_in_buffer = 0;
bn->seqinsert = 0;
bn->optimized_for_upgrade = 0;
bn->max_msn_applied = ZERO_MSN;
return bn;
}
@ -1017,6 +1016,7 @@ setup_brtnode_partitions(BRTNODE node, struct brtnode_fetch_extra* bfe) {
for (int i = 0; i < node->n_children; i++) {
BP_INIT_UNTOUCHED_CLOCK(node,i);
BP_STATE(node,i) = toku_brtnode_partition_state(bfe, i);
BP_WORKDONE(node,i) = 0;
if (BP_STATE(node,i) == PT_AVAIL) {
setup_available_brtnode_partition(node, i);
BP_TOUCH_CLOCK(node,i);

75
newbrt/brt.c
View file

@ -196,13 +196,22 @@ nonleaf_node_is_gorged (BRTNODE node) {
bool buffers_are_empty = TRUE;
toku_assert_entire_node_in_memory(node);
//
// the nonleaf node is gorged if the following holds true:
// - the buffers are non-empty
// - the total workdone by the buffers PLUS the size of the buffers
// is greater than node->nodesize (which as of Maxwell should be
// 4MB)
//
assert(node->height > 0);
for (int child = 0; child < node->n_children; ++child) {
size += BP_WORKDONE(node, child);
}
for (int child = 0; child < node->n_children; ++child) {
if (BNC_NBYTESINBUF(node, child) > 0) {
buffers_are_empty = FALSE;
break;
}
size += BP_WORKDONE(node, child);
}
return ((size > node->nodesize)
&&
@ -255,7 +264,7 @@ static u_int32_t compute_child_fullhash (CACHEFILE cf, BRTNODE node, int childnu
abort(); return 0;
}
static void maybe_apply_ancestors_messages_to_node (BRT t, BRTNODE node, ANCESTORS ancestors, struct pivot_bounds const * const bounds, bool *made_change);
static void maybe_apply_ancestors_messages_to_node (BRT t, BRTNODE node, ANCESTORS ancestors, struct pivot_bounds const * const bounds);
static long brtnode_memory_size (BRTNODE node);
@ -281,8 +290,7 @@ int toku_pin_brtnode (BRT brt, BLOCKNUM blocknum, u_int32_t fullhash,
unlockers);
if (r==0) {
BRTNODE node = node_v;
bool made_change;
maybe_apply_ancestors_messages_to_node(brt, node, ancestors, bounds, &made_change);
maybe_apply_ancestors_messages_to_node(brt, node, ancestors, bounds);
*node_p = node;
// printf("%*sPin %ld\n", 8-node->height, "", blocknum.b);
} else {
@ -313,8 +321,7 @@ void toku_pin_brtnode_holding_lock (BRT brt, BLOCKNUM blocknum, u_int32_t fullha
);
assert(r==0);
BRTNODE node = node_v;
bool made_change;
maybe_apply_ancestors_messages_to_node(brt, node, ancestors, bounds, &made_change);
maybe_apply_ancestors_messages_to_node(brt, node, ancestors, bounds);
*node_p = node;
}
@ -1614,7 +1621,7 @@ struct setval_extra_s {
u_int32_t idx;
LEAFENTRY le;
TOKULOGGER logger;
int made_change;
bool made_change;
uint64_t * workdonep; // set by brt_leaf_apply_cmd_once()
};
@ -1723,7 +1730,7 @@ static int do_update(BRT t, BASEMENTNODE bn, SUBTREE_EST se, BRT_MSG cmd, int id
}
// should be static, but used by test program(s)
void
static void
brt_leaf_put_cmd (
BRT t,
BASEMENTNODE bn,
@ -1738,13 +1745,6 @@ brt_leaf_put_cmd (
{
uint64_t workdone_total = 0; // may be for one row or for many (or all) rows in leaf (if broadcast message)
// ignore messages that have already been applied to this leaf
if (cmd->msn.msn <= bn->max_msn_applied.msn) {
// TODO3514 add accountability counter here
goto exit;
}
else bn->max_msn_applied = cmd->msn;
TOKULOGGER logger = toku_cachefile_logger(t->cf);
LEAFENTRY storeddata;
@ -1951,10 +1951,8 @@ brt_leaf_put_cmd (
// node->dirty = 1;
exit:
if (workdonep)
*workdonep = workdone_total;
VERIFY_NODE(t, node);
return;
}
@ -4922,8 +4920,8 @@ apply_buffer_messages_to_basement_node (
SUBTREE_EST se,
BRTNODE ancestor,
int childnum,
struct pivot_bounds const * const bounds,
bool *made_change
MSN min_applied_msn,
struct pivot_bounds const * const bounds
)
// Effect: For each messages in ANCESTOR that is between lower_bound_exclusive (exclusive) and upper_bound_inclusive (inclusive), apply the message to the node.
// In ANCESTOR, the relevant messages are all in the buffer for child number CHILDNUM.
@ -4957,14 +4955,15 @@ apply_buffer_messages_to_basement_node (
({
DBT hk;
toku_fill_dbt(&hk, key, keylen);
if (msn.msn > bn->max_msn_applied.msn && (!msg_type_has_key(type) || key_is_in_leaf_range(t, &hk, lbe_ptr, ubi_ptr))) {
if (msn.msn > min_applied_msn.msn && (!msg_type_has_key(type) || key_is_in_leaf_range(t, &hk, lbe_ptr, ubi_ptr))) {
DBT hv;
BRT_MSG_S brtcmd = { (enum brt_msg_type)type, msn, xids, .u.id = {&hk,
toku_fill_dbt(&hv, val, vallen)} };
uint64_t workdone_this_leaf = 0;
bool made_change;
brt_leaf_put_cmd(t,
bn, se,
&brtcmd, made_change, &workdone_this_leaf);
&brtcmd, &made_change, &workdone_this_leaf);
BP_WORKDONE(ancestor, childnum) += workdone_this_leaf;
workdone_this_leaf_total += workdone_this_leaf;
}
@ -4977,6 +4976,7 @@ apply_buffer_messages_to_basement_node (
return r;
}
/*
//###########
static void
maybe_flush_pinned_node(BRT t, BRTNODE node, int childnum, BRTNODE child) {
@ -5065,11 +5065,13 @@ maybe_flush_pinned_node(BRT t, BRTNODE node, int childnum, BRTNODE child) {
}
}
*/
/*
static void
apply_ancestors_messages_to_leafnode_and_maybe_flush (BRT t, BASEMENTNODE bm, SUBTREE_EST se, ANCESTORS ancestors, BRTNODE child,
const struct pivot_bounds const *bounds, bool *made_change)
MSN min_applied_msn,
const struct pivot_bounds const *bounds)
// Effect: Go through ancestors list applying messages from first ancestor (height one), then next, until
// all messages have been applied.
// Then mark the node as up_to_date.
@ -5086,17 +5088,18 @@ apply_ancestors_messages_to_leafnode_and_maybe_flush (BRT t, BASEMENTNODE bm, SU
// With background flushing we will be able to back to a simpler loop (since the recursion will be tail recursion).
{
if (ancestors) {
apply_buffer_messages_to_basement_node(t, bm, se, ancestors->node, ancestors->childnum, bounds, made_change);
apply_ancestors_messages_to_leafnode_and_maybe_flush(t, bm, se, ancestors->next, ancestors->node, bounds, made_change);
apply_buffer_messages_to_basement_node(t, bm, se, ancestors->node, ancestors->childnum, min_applied_msn, bounds);
apply_ancestors_messages_to_leafnode_and_maybe_flush(t, bm, se, ancestors->next, ancestors->node, min_applied_msn, bounds);
maybe_flush_pinned_node(t, ancestors->node, ancestors->childnum, child);
} else {
// have just applied messages stored in root
bm->soft_copy_is_up_to_date = true;
}
}
*/
static void
maybe_apply_ancestors_messages_to_node (BRT t, BRTNODE node, ANCESTORS ancestors, struct pivot_bounds const * const bounds, bool *made_change)
maybe_apply_ancestors_messages_to_node (BRT t, BRTNODE node, ANCESTORS ancestors, struct pivot_bounds const * const bounds)
// Effect:
// Bring a leaf node up-to-date according to all the messages in the ancestors.
// If the leaf node is already up-to-date then do nothing.
@ -5119,22 +5122,20 @@ maybe_apply_ancestors_messages_to_node (BRT t, BRTNODE node, ANCESTORS ancestors
SUBTREE_EST curr_se = &BP_SUBTREE_EST(node,i);
ANCESTORS curr_ancestors = ancestors;
struct pivot_bounds curr_bounds = next_pivot_keys(node, i, bounds);
BRTNODE child = node;
while (curr_ancestors) {
height++;
apply_ancestors_messages_to_leafnode_and_maybe_flush(
t,
curr_bn,
curr_se,
curr_ancestors,
child,
&curr_bounds,
made_change
);
apply_buffer_messages_to_basement_node(
t,
curr_bn,
curr_se,
curr_ancestors->node,
curr_ancestors->childnum,
node->max_msn_applied_to_node_on_disk,
&curr_bounds
);
if (curr_ancestors->node->max_msn_applied_to_node_in_memory.msn > node->max_msn_applied_to_node_in_memory.msn) {
node->max_msn_applied_to_node_in_memory = curr_ancestors->node->max_msn_applied_to_node_in_memory;
}
child = curr_ancestors->node;
curr_ancestors= curr_ancestors->next;
}
BLB_SOFTCOPYISUPTODATE(node, i) = TRUE;