robj *lookupKey(redisDb *db, robj *key) {
dictEntry *de = dictFind(db->dict,key->ptr);
if (de) {
robj *val = dictGetEntryVal(de);
if (server.vm_enabled) {
if (val->storage == REDIS_VM_MEMORY ||
val->storage == REDIS_VM_SWAPPING)
{
/* If we were swapping the object out, cancel the operation */
if (val->storage == REDIS_VM_SWAPPING)
vmCancelThreadedIOJob(val);
/* Update the access time for the aging algorithm. */
val->lru = server.lruclock;
} else {
int notify = (val->storage == REDIS_VM_LOADING);
/* Our value was swapped on disk. Bring it at home. */
redisAssert(val->type == REDIS_VMPOINTER);
val = vmLoadObject(val);
dictGetEntryVal(de) = val;
/* Clients blocked by the VM subsystem may be waiting for
* this key... */
if (notify) handleClientsBlockedOnSwappedKey(db,key);
}
}
return val;
} else {
return NULL;
}
}
static robj *lookupKey(redisDb *db, robj *key) {
dictEntry *de = dictFind(db->dict,key);
if (de) {
robj *key = dictGetEntryKey(de);
robj *val = dictGetEntryVal(de);
if (server.vm_enabled) {
if (key->storage == REDIS_VM_MEMORY ||
key->storage == REDIS_VM_SWAPPING)
{
/* If we were swapping the object out, stop it, this key
* was requested. */
if (key->storage == REDIS_VM_SWAPPING)
vmCancelThreadedIOJob(key);
/* Update the access time of the key for the aging algorithm. */
key->vm.atime = server.unixtime;
} else {
int notify = (key->storage == REDIS_VM_LOADING);
/* Our value was swapped on disk. Bring it at home. */
redisAssert(val == NULL);
val = vmLoadObject(key);
dictGetEntryVal(de) = val;
/* Clients blocked by the VM subsystem may be waiting for
* this key... */
if (notify) handleClientsBlockedOnSwappedKey(db,key);
}
}
return val;
} else {
return NULL;
}
}
/* Delete a key, value, and associated expiration entry if any, from the DB */
int dbDelete(redisDb *db, robj *key) {
/* If VM is enabled make sure to awake waiting clients for this key:
* deleting the key will kill the I/O thread bringing the key from swap
* to memory, so the client will never be notified and unblocked if we
* don't do it now. */
if (server.vm_enabled) handleClientsBlockedOnSwappedKey(db,key);
/* Deleting an entry from the expires dict will not free the sds of
* the key, because it is shared with the main dictionary. */
if (dictSize(db->expires) > 0) dictDelete(db->expires,key->ptr);
return dictDelete(db->dict,key->ptr) == DICT_OK;
}
/* Delete a key, value, and associated expiration entry if any, from the DB */
int dbDelete(redisDb *db, robj *key) {
/* If diskstore is enabled make sure to awake waiting clients for this key
* as it is not really useful to wait for a key already deleted to be
* loaded from disk. */
if (server.ds_enabled) {
handleClientsBlockedOnSwappedKey(db,key);
cacheSetKeyDoesNotExist(db,key);
}
/* Deleting an entry from the expires dict will not free the sds of
* the key, because it is shared with the main dictionary. */
if (dictSize(db->expires) > 0) dictDelete(db->expires,key->ptr);
return dictDelete(db->dict,key->ptr) == DICT_OK;
}
robj *lookupKey(redisDb *db, robj *key) {
dictEntry *de = dictFind(db->dict,key->ptr);
if (de) {
robj *val = dictGetEntryVal(de);
/* Update the access time for the aging algorithm.
* Don't do it if we have a saving child, as this will trigger
* a copy on write madness. */
if (server.bgsavechildpid == -1 && server.bgrewritechildpid == -1)
val->lru = server.lruclock;
if (server.vm_enabled) {
if (val->storage == REDIS_VM_MEMORY ||
val->storage == REDIS_VM_SWAPPING)
{
/* If we were swapping the object out, cancel the operation */
if (val->storage == REDIS_VM_SWAPPING)
vmCancelThreadedIOJob(val);
} else {
int notify = (val->storage == REDIS_VM_LOADING);
/* Our value was swapped on disk. Bring it at home. */
redisAssert(val->type == REDIS_VMPOINTER);
val = vmLoadObject(val);
dictGetEntryVal(de) = val;
/* Clients blocked by the VM subsystem may be waiting for
* this key... */
if (notify) handleClientsBlockedOnSwappedKey(db,key);
}
}
server.stat_keyspace_hits++;
return val;
} else {
server.stat_keyspace_misses++;
return NULL;
}
}
/* Every time a thread finished a Job, it writes a byte into the write side
* of an unix pipe in order to "awake" the main thread, and this function
* is called.
*
* Note that this is called both by the event loop, when a I/O thread
* sends a byte in the notification pipe, and is also directly called from
* waitEmptyIOJobsQueue().
*
* In the latter case we don't want to swap more, so we use the
* "privdata" argument setting it to a not NULL value to signal this
* condition. */
void vmThreadedIOCompletedJob(aeEventLoop *el, int fd, void *privdata,
int mask)
{
char buf[1];
int retval, processed = 0, toprocess = -1, trytoswap = 1;
REDIS_NOTUSED(el);
REDIS_NOTUSED(mask);
REDIS_NOTUSED(privdata);
if (privdata != NULL) trytoswap = 0; /* check the comments above... */
/* For every byte we read in the read side of the pipe, there is one
* I/O job completed to process. */
#ifndef _WIN32
while((retval = read(fd,buf,1)) == 1) {
#else
DWORD pipe_is_on = 0;
while (1) {
retval = 0;
/*Windows fix: We need to peek pipe, since read would block. */
if (!PeekNamedPipe((HANDLE) _get_osfhandle(fd), NULL, 0, NULL, &pipe_is_on, NULL)) {
redisLog(REDIS_DEBUG,"PeekReadPipe failed %s", strerror(GetLastError()));
break;
}
/* No data on pipe */
if (!pipe_is_on)
break;
if ((retval = read(fd,buf,1)) != 1)
break;
#endif
iojob *j;
listNode *ln;
struct dictEntry *de;
/* Get the processed element (the oldest one) */
lockThreadedIO();
redisLog(REDIS_DEBUG,"Processing I/O completed job");
redisAssert(listLength(server.io_processed) != 0);
if (toprocess == -1) {
toprocess = (listLength(server.io_processed)*REDIS_MAX_COMPLETED_JOBS_PROCESSED)/100;
if (toprocess <= 0) toprocess = 1;
}
ln = listFirst(server.io_processed);
j = ln->value;
listDelNode(server.io_processed,ln);
unlockThreadedIO();
/* If this job is marked as canceled, just ignore it */
if (j->canceled) {
freeIOJob(j);
continue;
}
/* Post process it in the main thread, as there are things we
* can do just here to avoid race conditions and/or invasive locks */
redisLog(REDIS_DEBUG,"COMPLETED Job type: %d, ID %p, key: %s", j->type, (void*)j->id, (unsigned char*)j->key->ptr);
de = dictFind(j->db->dict,j->key->ptr);
redisAssert(de != NULL);
if (j->type == REDIS_IOJOB_LOAD) {
redisDb *db;
vmpointer *vp = dictGetEntryVal(de);
/* Key loaded, bring it at home */
vmMarkPagesFree(vp->page,vp->usedpages);
redisLog(REDIS_DEBUG, "VM: object %s loaded from disk (threaded)",
(unsigned char*) j->key->ptr);
server.vm_stats_swapped_objects--;
server.vm_stats_swapins++;
dictGetEntryVal(de) = j->val;
incrRefCount(j->val);
db = j->db;
/* Handle clients waiting for this key to be loaded. */
handleClientsBlockedOnSwappedKey(db,j->key);
freeIOJob(j);
zfree(vp);
} else if (j->type == REDIS_IOJOB_PREPARE_SWAP) {
/* Now we know the amount of pages required to swap this object.
* Let's find some space for it, and queue this task again
* rebranded as REDIS_IOJOB_DO_SWAP. */
if (!vmCanSwapOut() ||
vmFindContiguousPages(&j->page,j->pages) == REDIS_ERR)
{
/* Ooops... no space or we can't swap as there is
* a fork()ed Redis trying to save stuff on disk. */
j->val->storage = REDIS_VM_MEMORY; /* undo operation */
freeIOJob(j);
} else {
/* Note that we need to mark this pages as used now,
* if the job will be canceled, we'll mark them as freed
* again. */
vmMarkPagesUsed(j->page,j->pages);
j->type = REDIS_IOJOB_DO_SWAP;
lockThreadedIO();
queueIOJob(j);
unlockThreadedIO();
}
} else if (j->type == REDIS_IOJOB_DO_SWAP) {
vmpointer *vp;
/* Key swapped. We can finally free some memory. */
if (j->val->storage != REDIS_VM_SWAPPING) {
vmpointer *vp = (vmpointer*) j->id;
printf("storage: %d\n",vp->storage);
printf("key->name: %s\n",(char*)j->key->ptr);
printf("val: %p\n",(void*)j->val);
printf("val->type: %d\n",j->val->type);
printf("val->ptr: %s\n",(char*)j->val->ptr);
}
redisAssert(j->val->storage == REDIS_VM_SWAPPING);
vp = createVmPointer(j->val);
vp->page = j->page;
vp->usedpages = j->pages;
dictGetEntryVal(de) = vp;
/* Fix the storage otherwise decrRefCount will attempt to
* remove the associated I/O job */
j->val->storage = REDIS_VM_MEMORY;
decrRefCount(j->val);
redisLog(REDIS_DEBUG,
"VM: object %s swapped out at %lld (%lld pages) (threaded)",
(unsigned char*) j->key->ptr,
(unsigned long long) j->page, (unsigned long long) j->pages);
server.vm_stats_swapped_objects++;
server.vm_stats_swapouts++;
freeIOJob(j);
/* Put a few more swap requests in queue if we are still
* out of memory */
if (trytoswap && vmCanSwapOut() &&
zmalloc_used_memory() > server.vm_max_memory)
{
int more = 1;
while(more) {
lockThreadedIO();
more = listLength(server.io_newjobs) <
(unsigned) server.vm_max_threads;
unlockThreadedIO();
/* Don't waste CPU time if swappable objects are rare. */
if (vmSwapOneObjectThreaded() == REDIS_ERR) {
trytoswap = 0;
break;
}
}
}
}
processed++;
if (processed == toprocess) return;
}
if (retval < 0 && errno != EAGAIN) {
redisLog(REDIS_WARNING,
"WARNING: read(2) error in vmThreadedIOCompletedJob() %s",
strerror(errno));
}
}
void lockThreadedIO(void) {
pthread_mutex_lock(&server.io_mutex);
}
