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:4372] updating system comments on top of brt.c, some comments too

Browse source 
git-svn-id: file:///svn/toku/tokudb@39179 c7de825b-a66e-492c-adef-691d508d4ae1
This commit is contained in:
John Esmet 2013-04-17 00:00:07 -04:00 • committed by Yoni Fogel
parent e1fcefad24
commit b515867acc
2 changed files with 12 additions and 16 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

27
newbrt/brt.c
View file

@ -10,16 +10,17 @@ Managing the tree shape: How insertion, deletion, and querying work
When we insert a message into the BRT, here's what happens.
Insert_a_message_at_root (msg)
root = find the root
insert_the_message_into_the_buffers_of(msg, root)
If the root is way too full then process the root ourself. "Way too full" means something like twice as much messages as it's supposed to have.
else If the root needs to be split, then split it
else if the root's buffers are too full then (it must be a nonleaf)
create a work item to process the root. The workitem specifies a height and a key (the height is the height of the root, and the key can be any key)
endif
If the brt file is fragmented, and the file isn't being shrunk, then set file->being_shrunk and schedule a work item to shrink the file.
insert_a_message_at_root(msg):
root = find the root node
if root needs to be split (has fissible reactivity):
split the root
root = find the new root node
insert_msg_into_node_buffer(root, msg)
if root has too many messages in its buffer and is a nonleaf node:
flush the buffer on a background thread
flush_nonleaf_node(node, height, key):
To process a nonleaf node (height, key)
Note: Height is always > 0.
Note: This process occurs asynchrnously, but we get the YDB lock at the beginning.
@ -90,7 +91,6 @@ Split_or_merge (node, childnum) {
}
}
Lookup:
As of #3312, we don't do any tree shaping on lookup.
We don't promote eagerly or use aggressive promotion or passive-aggressive promotion. We just push messages down according to the traditional BRT algorithm
@ -238,6 +238,7 @@ toku_brt_nonleaf_is_gorged (BRTNODE node) {
(!buffers_are_empty));
}
// FIXME this is not used
static inline void add_to_brt_status(u_int64_t* val, u_int64_t data) {
(*val) += data;
}
@ -507,7 +508,6 @@ update_header_stats(STAT64INFO headerstats, STAT64INFO delta) {
(void) __sync_fetch_and_add(&(headerstats->numbytes), delta->numbytes);
}
// This is the ONLY place where a node is marked as dirty, other than toku_initialize_empty_brtnode().
void
toku_mark_node_dirty(BRTNODE node) {
@ -528,8 +528,6 @@ toku_mark_node_dirty(BRTNODE node) {
node->dirty = 1;
}
//fd is protected (must be holding fdlock)
void toku_brtnode_flush_callback (CACHEFILE cachefile, int fd, BLOCKNUM nodename, void *brtnode_v, void *extraargs, PAIR_ATTR size __attribute__((unused)), PAIR_ATTR* new_size, BOOL write_me, BOOL keep_me, BOOL for_checkpoint) {
struct brt_header *h = extraargs;
@ -616,7 +614,6 @@ int toku_brtnode_fetch_callback (CACHEFILE UU(cachefile), int fd, BLOCKNUM noden
return r;
}
void toku_brtnode_pe_est_callback(
void* brtnode_pv,
long* bytes_freed_estimate,
@ -663,7 +660,6 @@ exit:
return;
}
static void
compress_internal_node_partition(BRTNODE node, int i)
{
@ -2225,7 +2221,6 @@ void toku_apply_cmd_to_leaf(
VERIFY_NODE(t, node);
}
static void push_something_at_root (BRT brt, BRTNODE *nodep, BRT_MSG cmd)
// Effect: Put CMD into brt's root node, and update
// the value of root's max_msn_applied_to_node_on_disk

1
newbrt/cachetable.c
View file

@ -1291,6 +1291,7 @@ static void cachetable_fetch_pair(
void *toku_value = 0;
PAIR_ATTR attr;
// FIXME this should be enum cachetable_dirty, right?
int dirty = 0;
WHEN_TRACE_CT(printf("%s:%d CT: fetch_callback(%lld...)\n", __FILE__, __LINE__, key));