Commit f5c8ab0a authored by Zardosht Kasheff's avatar Zardosht Kasheff Committed by Yoni Fogel

[t:3997], add some comments to cachetable functions that will inevitabely be...

[t:3997], add some comments to cachetable functions that will inevitabely be tricky when dealing with cache pressure

git-svn-id: file:///svn/toku/tokudb@35997 c7de825b-a66e-492c-adef-691d508d4ae1
parent 05fe2780
......@@ -2585,13 +2585,13 @@ static int cachetable_flush_cachefile(CACHETABLE ct, CACHEFILE cf) {
workqueue_init(&cq);
// find all of the pairs owned by a cachefile and redirect their completion
// to a completion queue. flush and remove pairs in the IDLE state if they
// are dirty. pairs in the READING or WRITING states are already in the
// work queue.
// to a completion queue. If an unlocked PAIR is dirty, flush and remove
// the PAIR. Locked PAIRs are on either a reader/writer thread (or maybe a
// checkpoint thread?) and therefore will be placed on the completion queue.
// The assumptions above lead to this reasoning. All pairs belonging to
// this cachefile are either:
// - unlocked
// - locked and on a writer thread.
// - locked and on a writer thread (or possibly on a checkpoint thread?).
// We find all the pairs owned by the cachefile and do the following:
// - if the PAIR is clean and unlocked, then remove the PAIR
// - if the PAIR is dirty and unlocked, write the PAIR to disk on a writer thread
......@@ -2600,7 +2600,7 @@ static int cachetable_flush_cachefile(CACHETABLE ct, CACHEFILE cf) {
// when the function started).
// - Once the writer thread is done with a PAIR, remove it
//
// An question from Zardosht: Is it possible for a checkpoint thread
// A question from Zardosht: Is it possible for a checkpoint thread
// to be running, and also trying to get access to a PAIR while
// the PAIR is on the writer thread? Will this cause problems?
// This question is encapsulated in #3941
......@@ -2653,7 +2653,7 @@ static int cachetable_flush_cachefile(CACHETABLE ct, CACHEFILE cf) {
p->cq = &cq;
//
// Once again, the assumption is that any PAIR
// that is either unlocked or on a writer thread work queue
// is either unlocked or on a writer thread work queue
//
if (!nb_mutex_writers(&p->nb_mutex)) {
flush_and_maybe_remove(ct, p);
......@@ -2664,6 +2664,22 @@ static int cachetable_flush_cachefile(CACHETABLE ct, CACHEFILE cf) {
toku_free(list);
// wait for all of the pairs in the work queue to complete
//
// An important assumption here is that none of the PAIRs that we
// pop off the work queue need to be written out to disk. So, it is
// safe to simply call cachetable_maybe_remove_and_free_pair on
// the PAIRs we find. The reason we can make this assumption
// is based on the assumption that upon entry of this function,
// all PAIRs belonging to this cachefile are either idle,
// being processed by a writer thread, or being processed by a kibbutz.
// At this point in the code, kibbutz work is finished and we
// assume the client will not add any more kibbutz work for this cachefile.
//
// If it were possible
// for some thread to change the state of the node before passing
// it off here, and a write to disk were necessary, then the code
// below would be wrong.
//
for (i=0; i<nfound; i++) {
cachetable_unlock(ct);
WORKITEM wi = 0;
......@@ -2746,7 +2762,21 @@ int toku_cachetable_unpin_and_remove (CACHEFILE cachefile, CACHEKEY key, BOOL ct
assert(nb_mutex_writers(&p->nb_mutex));
nb_mutex_write_unlock(&p->nb_mutex);
//
// need to find a way to assert that ONLY the checkpoint thread may be blocked here
// need to find a way to assert that
// ONLY the checkpoint thread may be blocked here
//
// The assumption here is that only the checkpoint thread may
// be blocked here. No writer thread may be a blocked writer,
// because the writer thread has only locked PAIRs.
// The writer thread does not try to acquire a lock. It cannot be a
// client thread either, because no client thread should be trying
// to lock a node that another thread is trying to remove
// from the cachetable. It cannot be a kibbutz thread either
// because the client controls what work is done on the kibbutz,
// and should be smart enough to make sure that no other thread
// tries to lock a PAIR while trying to unpin_and_remove it. So,
// the only thread that is left that can possibly be a blocked
// writer is the checkpoint thread.
//
if (nb_mutex_blocked_writers(&p->nb_mutex)>0) {
struct workqueue cq;
......@@ -2776,6 +2806,10 @@ int toku_cachetable_unpin_and_remove (CACHEFILE cachefile, CACHEKEY key, BOOL ct
cachetable_lock(ct);
assert(nb_mutex_writers(&p->nb_mutex) == 1);
BOOL destroyed = FALSE;
// Because we assume it is just the checkpoint thread
// that may have been blocked (as argued above),
// it is safe to simply remove the PAIR from the
// cachetable. We don't need to write anything out.
cachetable_complete_write_pair(ct, p, TRUE, &destroyed);
if (destroyed) {
break;
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment