int ea_delete(EA *ea)
{
int rc = 0;
if ( ea->st_type == EA_STORE_TYPE_MEMORY )
{
KV_TRC(pAT, "free ea %p ea->st_memory %p", ea, ea->st_memory);
// Simple free the block of store
if (ea->st_memory)
{
free(ea->st_memory);
ea->st_memory=NULL;
}
}
else
{
// Call to close out the chunk and free the space
// for the device name
rc = cblk_close(ea->st_flash, 0);
am_free(ea->st_device);
}
if ( rc == 0 )
{
KV_TRC(pAT, "free ea %p", ea);
am_free(ea);
}
return rc;
}
/**
*******************************************************************************
* \brief
******************************************************************************/
uint32_t kv_async_init_perf_io(uint32_t num_ctxt,
uint32_t jobs,
uint32_t npool,
uint32_t klen,
uint32_t vlen,
uint32_t LEN,
uint32_t secs)
{
uint32_t job = 0;
uint32_t ctxt = 0;
kv_t *db[jobs];
printf("ASYNC %dx%dx%d", klen, vlen, LEN); fflush(stdout);
for (job=0; job<jobs; job++)
{
db[job] = (kv_t*)kv_db_create_fixed(LEN, klen+job, vlen);
}
printf("."); fflush(stdout);
for (ctxt=0; ctxt<num_ctxt; ctxt++)
{
if (kv_async_init_ctxt_perf(ctxt, npool, secs)) return ENOMEM;
for (job=0; job<jobs; job++)
{
kv_async_set_job(KV_ASYNC_CB_SGD |
KV_ASYNC_CB_MULTI_CTXT_IO |
KV_ASYNC_CB_GTEST,
ctxt, job, db[job], vlen, LEN);
KV_TRC(pFT, "CREATE_JOB FIXED %dx%dx%d", klen+job, vlen, LEN);
}
}
printf("> "); fflush(stdout);
return 0;
}
/**
*******************************************************************************
* \brief
******************************************************************************/
void kv_async_init_ark_io(uint32_t num_ctxt,
uint32_t jobs,
uint32_t vlen,
uint32_t secs)
{
uint32_t klen = 16;
uint32_t LEN = 50;
uint32_t job = 0;
uint32_t ctxt = 0;
kv_t *db[jobs];
printf("ctxt:%d jobs:%d ASYNC %dx%dx%d",
num_ctxt, jobs, klen, vlen, LEN); fflush(stdout);
for (job=0; job<jobs; job++)
{
db[job] = (kv_t*)kv_db_create_fixed(LEN, klen+job, vlen);
}
printf("."); fflush(stdout);
for (ctxt=0; ctxt<num_ctxt; ctxt++)
{
kv_async_init_ctxt(ctxt, secs);
for (job=0; job<jobs; job++)
{
kv_async_set_job(KV_ASYNC_CB_SGD | KV_ASYNC_CB_MULTI_CTXT_IO,
ctxt, job, db[job], vlen, LEN);
KV_TRC(pFT, "CREATE_JOB FIXED %dx%dx%d", klen+job, vlen, LEN);
}
}
printf("> "); fflush(stdout);
}
/**
*******************************************************************************
* \brief
******************************************************************************/
void kv_async_init_ctxt(uint32_t ctxt, uint32_t secs)
{
char *env_FVT = getenv("FVT_DEV");
async_context_t *pCT = pCTs+ctxt;
if (ctxt < 0 || ctxt > KV_ASYNC_MAX_CONTEXTS)
{
printf("FFDC: kv_async_init_ctxt %d %X\n", ctxt, ctxt);
return;
}
memset(pCT, 0, sizeof(async_context_t));
memset(pCT->pCBs, 0, sizeof(async_CB_t)*KV_ASYNC_JOB_Q);
ASSERT_EQ(0, ark_create_verbose(env_FVT, &pCT->ark,
1048576,
4096,
1048576,
20,
256,
8*1024,
ARK_KV_VIRTUAL_LUN));
ASSERT_TRUE(NULL != pCT->ark);
pCT->flags |= KV_ASYNC_CT_RUNNING;
pCT->secs = secs;
KV_TRC(pFT, "init_ctxt ctxt:%d ark:%p secs:%d", ctxt, pCT->ark, pCT->secs);
}
/**
*******************************************************************************
* \brief
******************************************************************************/
void kv_async_set_job(uint32_t flags,
uint32_t ctxt,
uint32_t job,
kv_t *db,
uint32_t vlen,
uint32_t len)
{
async_context_t *pCT = pCTs+ctxt;
async_CB_t *pCB = NULL;
char type[4] = {0};
ASSERT_TRUE(NULL != pCT);
ASSERT_TRUE(ctxt >= 0);
ASSERT_TRUE(ctxt < KV_ASYNC_MAX_CONTEXTS);
ASSERT_TRUE(0 != len);
if (flags & KV_ASYNC_CB_SGD) sprintf(type, "SGD");
else sprintf(type, "REP");
pCB = pCT->pCBs+job; memset(pCB, 0, sizeof(async_CB_t));
pCB->ark = pCT->ark;
pCB->flags = flags | KV_ASYNC_CB_SET | KV_ASYNC_CB_QUEUED;
pCB->db = db;
pCB->regen = kv_db_fixed_regen_values;
pCB->len = len;
pCB->cb = kv_async_cb;
pCB->regen_len = vlen;
pCB->gvalue = (char*)malloc(pCB->regen_len);
pCB->b_mark = B_MARK;
pCB->e_mark = E_MARK;
KV_TRC(pFT, "CREATE_JOB: ctxt:%d ark:%p %s: pCB:%p flags:%X",
ctxt, pCT->ark, type, pCB, pCB->flags);
}
/**
*******************************************************************************
* \brief
******************************************************************************/
IV *iv_new(uint64_t n, uint64_t m) {
uint64_t bits = n * m;
uint64_t words = divup(bits, 64);
uint64_t bytes = sizeof(IV) + words * sizeof(uint64_t);
IV *iv = am_malloc(bytes);
if (iv == NULL)
{
errno = ENOMEM;
KV_TRC_FFDC(pAT, "FFDC: n %"PRIu64" m %"PRIu64", errno = %d", n, m, errno);
}
else
{
memset(iv,0x00, bytes);
iv->n = n;
iv->m = m;
iv->bits = bits;
iv->words = words;
iv->mask = 1;
iv->mask <<= m;
iv->mask -= 1;
iv->bar = 64 - m;
}
KV_TRC(pAT, "iv:%p n:%ld m:%ld", iv, n, m);
return iv;
}
/**
*******************************************************************************
* \brief
******************************************************************************/
uint32_t kv_async_init_ctxt_perf(uint32_t ctxt, uint32_t npool, uint32_t secs)
{
char *env_FVT = getenv("FVT_DEV");
async_context_t *pCT = pCTs+ctxt;
if (ctxt < 0 || ctxt > KV_ASYNC_MAX_CONTEXTS)
{
printf("FFDC: kv_async_init_ctxt %d %X\n", ctxt, ctxt);
return EINVAL;
}
memset(pCT, 0, sizeof(async_context_t));
memset(pCT->pCBs, 0, sizeof(async_CB_t)*KV_ASYNC_JOB_Q);
if (ark_create_verbose(env_FVT, &pCT->ark,
1048576,
4096,
1048576,
npool,
256,
8*1024,
ARK_KV_VIRTUAL_LUN))
{
printf("ark_create failed for ctxt:%d\n", ctxt);
return ENOMEM;
}
if (NULL == pCT->ark) return ENOMEM;
pCT->flags |= KV_ASYNC_CT_RUNNING;
pCT->secs = secs;
KV_TRC(pFT, "init_ctxt ctxt:%d ark:%p", ctxt, pCT->ark);
return 0;
}
void ark_persistence_calc(_ARK *_arkp)
{
uint64_t bytes = 0;
// We need to determine the total size of the data
// that needs to be persisted. Items that we persist
// are:
//
// - Configuration
// - Hash Table (hash_t)
// - Block List (BL)
// - IV->data
// Configuration
_arkp->pers_cs_bytes = sizeof(p_cntr_t) + sizeof(P_ARK_t);
// Hash Table
_arkp->pers_cs_bytes += sizeof(hash_t) + (_arkp->ht->n * sizeof(uint64_t));
// Block List
_arkp->pers_cs_bytes += sizeof(BL);
// calculate for a full ark
bytes = _arkp->pers_cs_bytes;
bytes += _arkp->bl->list->words * sizeof(uint64_t);
_arkp->pers_max_blocks = divup(bytes, _arkp->bsize);
KV_TRC(pAT, "PERSIST_CALC pers_cs_bytes:%ld pers_max_blocks:%ld",
_arkp->pers_cs_bytes, _arkp->pers_max_blocks);
return;
}
/**
*******************************************************************************
* \brief
******************************************************************************/
void kv_async_init_job(uint32_t flags,
uint32_t ctxt,
uint32_t job,
uint32_t klen,
uint32_t vlen,
uint32_t len)
{
kv_t *db = (kv_t*)kv_db_create_fixed(len, klen, vlen);
ASSERT_TRUE(NULL != db);
KV_TRC(pFT, "CREATE_JOB FIXED %dx%dx%d", klen, vlen, len);
kv_async_set_job(flags|KV_ASYNC_CB_GTEST, ctxt, job, db, vlen, len);
}
/**
*******************************************************************************
* \brief
******************************************************************************/
static void kv_async_perf_done(async_CB_t *pCB)
{
++pCB->perf_loops;
if (pCB->flags & KV_ASYNC_CB_SHUTDOWN)
{
KV_TRC(pFT, "shutdown %p %d", pCB, pCB->perf_loops);
pCB->flags &= ~(KV_ASYNC_CB_SET |
KV_ASYNC_CB_GET |
KV_ASYNC_CB_WRITE_PERF |
KV_ASYNC_CB_READ_PERF |
KV_ASYNC_CB_RUNNING);
return;
}
kv_async_dispatch(pCB);
}
/**
*******************************************************************************
* \brief
******************************************************************************/
void kv_async_run_jobs(void)
{
async_CB_t *pCB = NULL;
uint32_t ctxt_running = 0;
uint32_t jobs_running = 0;
uint32_t i = 0;
uint32_t next = 0;
uint32_t elapse = 0;
uint32_t inject = 0;
uint32_t secs = 0;
uint32_t log_interval = 600;
uint64_t ops = 0;
uint64_t ios = 0;
uint32_t tops = 0;
uint32_t tios = 0;
uint32_t perf = 0;
KV_TRC(pFT, "ASYNC START: 0 minutes");
if (!(pCTs->pCBs->flags & KV_ASYNC_CB_RUNNING)) start = time(0);
next = log_interval;
do
{
ctxt_running = FALSE;
if (elapse > next)
{
KV_TRC(pFT, "ASYNC RUNNING: %d elapsed minutes", elapse/60);
next += log_interval;
}
for (i=0; i<KV_ASYNC_MAX_CONTEXTS; i++)
{
if (! (pCTs[i].flags & KV_ASYNC_CT_RUNNING)) continue;
jobs_running = kv_async_dispatch_jobs(i);
if (!jobs_running)
{
pCTs[i].flags &= ~KV_ASYNC_CT_RUNNING;
pCTs[i].flags |= KV_ASYNC_CT_DONE;
KV_TRC(pFT, "ASYNC DONE ctxt %d %x", i, pCTs[i].flags);
continue;
}
else
{
ctxt_running = TRUE;
}
elapse = time(0) - start;
if (elapse >= inject &&
pCTs[i].flags & KV_ASYNC_CT_ERROR_INJECT)
{
KV_TRC_FFDC(pFT, "FFDC: INJECT ERRORS");
FVT_KV_INJECT_READ_ERROR;
FVT_KV_INJECT_WRITE_ERROR;
FVT_KV_INJECT_ALLOC_ERROR;
++inject;
}
if (elapse >= pCTs[i].secs)
{
for (pCB=pCTs[i].pCBs;pCB<pCTs[i].pCBs+KV_ASYNC_JOB_Q;pCB++)
{
if ((pCB->flags & KV_ASYNC_CB_RUNNING ||
pCB->flags & KV_ASYNC_CB_QUEUED)
&&
(!(pCB->flags & KV_ASYNC_CB_SHUTDOWN)) )
{
pCB->flags |= KV_ASYNC_CB_SHUTDOWN;
KV_TRC_IO(pFT, "SHUTDOWN pCB %p (%d >= %d)", pCB, elapse, pCTs[i].secs);
}
}
}
usleep(100);
}
}
while (ctxt_running);
stop = time(0);
secs = stop - start;
KV_TRC(pFT, "ASYNC RUNNING DONE: %d minutes", elapse/60);
/* log cleanup, since the first ark_delete closes the log file */
for (i=0; i<KV_ASYNC_MAX_CONTEXTS; i++)
{
if (pCTs[i].flags & KV_ASYNC_CT_DONE)
KV_TRC(pFT, "ASYNC CLEANUP: ctxt:%d ark:%p", i, pCTs[i].ark);
}
/* check for MULTI_CTXT_IO, destroy common kv dbs */
for (pCB=pCTs->pCBs;pCB<pCTs->pCBs+KV_ASYNC_JOB_Q;pCB++)
{
if (pCB->flags & KV_ASYNC_CB_MULTI_CTXT_IO)
{
kv_db_destroy(pCB->db, pCB->len);
}
}
for (i=0; i<KV_ASYNC_MAX_CONTEXTS; i++)
{
/* if this context didn't run any I/O */
if (! (pCTs[i].flags & KV_ASYNC_CT_DONE)) continue;
pCTs[i].flags &= ~KV_ASYNC_CT_DONE;
/* if perf then don't delete the ark here */
if (pCTs[i].flags & KV_ASYNC_CT_PERF)
{
perf = TRUE;
continue;
}
(void)ark_stats(pCTs[i].ark, &ops, &ios);
tops += (uint32_t)ops;
tios += (uint32_t)ios;
KV_TRC(pFT, "PERF ark%p ops:%"PRIu64" ios:%"PRIu64"", pCTs[i].ark, ops, ios);
EXPECT_EQ(0, ark_delete(pCTs[i].ark));
}
if (!perf)
{
tops = tops / secs;
tios = tios / secs;
printf("op/s:%d io/s:%d secs:%d\n", tops, tios, secs);
KV_TRC(pFT, "PERF op/s:%d io/s:%d secs:%d",
tops, tios, secs);
}
}
/**
*******************************************************************************
* \brief
******************************************************************************/
void kv_async_job_perf(uint32_t jobs, uint32_t klen, uint32_t vlen,uint32_t len)
{
long int wr_us = 0;
long int rd_us = 0;
long int mil = 1000000;
float wr_s = 0;
float rd_s = 0;
float wr_mb = 0;
float rd_mb = 0;
uint64_t mb64_1 = (uint64_t)KV_1M;
uint64_t wr_bytes = 0;
uint64_t rd_bytes = 0;
uint64_t ops = 0;
uint64_t post_ops = 0;
uint64_t ios = 0;
uint64_t post_ios = 0;
float wr_ops = 0;
float wr_ios = 0;
float rd_ops = 0;
float rd_ios = 0;
uint32_t secs = 5;
uint32_t job = 0;
struct timeval stop, start;
kv_async_init_ctxt(0, secs);
for (job=0; job<jobs; job++)
{
kv_async_init_job(KV_ASYNC_CB_SGD|KV_ASYNC_CB_WRITE_PERF,
ASYNC_SINGLE_CONTEXT,
job, klen+job, vlen, len);
}
pCTs->flags |= KV_ASYNC_CT_PERF;
/* do writes */
(void)ark_stats(kv_async_get_ark(ASYNC_SINGLE_CONTEXT), &ops, &ios);
KV_TRC(pFT, "PERF wr: ops:%"PRIu64" ios:%"PRIu64"", ops, ios);
gettimeofday(&start, NULL);
kv_async_run_jobs(); /* run write jobs */
KV_TRC(pFT, "writes done");
gettimeofday(&stop, NULL);
wr_us += (stop.tv_sec*mil + stop.tv_usec) -
(start.tv_sec*mil + start.tv_usec);
(void)ark_stats(kv_async_get_ark(ASYNC_SINGLE_CONTEXT),&post_ops,&post_ios);
KV_TRC(pFT, "PERF wr: ops:%"PRIu64" ios:%"PRIu64"", post_ops, post_ios);
wr_ops += post_ops - ops;
wr_ios += post_ios - ios;
/* calc bytes written */
for (job=0; job<jobs; job++)
{
wr_bytes += (klen+vlen+job)*len*(pCTs->pCBs+job)->perf_loops;
}
/* do reads */
for (job=0; job<jobs; job++)
{
if ((pCTs->pCBs+job)->perf_loops)
{
(pCTs->pCBs+job)->perf_loops = 0;
(pCTs->pCBs+job)->flags = KV_ASYNC_CB_GET |
KV_ASYNC_CB_QUEUED |
KV_ASYNC_CB_READ_PERF;
}
}
pCTs->flags |= KV_ASYNC_CT_RUNNING;
(void)ark_stats(kv_async_get_ark(0), &ops, &ios);
KV_TRC(pFT, "PERF rd: ops:%"PRIu64" ios:%"PRIu64"", ops, ios);
gettimeofday(&start, NULL);
kv_async_run_jobs(); /* run read jobs */
gettimeofday(&stop, NULL);
KV_TRC(pFT, "reads done");
rd_us += (stop.tv_sec*mil + stop.tv_usec) -
(start.tv_sec*mil + start.tv_usec);
(void)ark_stats(kv_async_get_ark(0), &post_ops, &post_ios);
KV_TRC(pFT, "PERF rd: ops:%"PRIu64" ios:%"PRIu64"", post_ops, post_ios);
rd_ops += post_ops - ops;
rd_ios += post_ios - ios;
ASSERT_EQ(0, ark_delete(pCTs->ark));
/* calc bytes read */
for (job=0; job<jobs; job++)
{
rd_bytes += vlen*len*(pCTs->pCBs+job)->perf_loops;
kv_db_destroy((pCTs->pCBs+job)->db, (pCTs->pCBs+job)->len);
if ((pCTs->pCBs+job)->gvalue)
free((pCTs->pCBs+job)->gvalue);
}
/* calc and print results */
wr_s = (float)((float)wr_us/(float)mil);
wr_mb = (float)((double)wr_bytes / (double)mb64_1);
rd_s = (float)((float)rd_us/(float)mil);
rd_mb = (float)((double)rd_bytes / (double)mb64_1);
printf("ASYNC %dx%dx%d writes: %.3d jobs %2.3f mb in %.1f secs at ",
klen, vlen, len, jobs, wr_mb, wr_s);
printf("%2.3f mbps, %6.0f op/s, %.0f io/s\n",
wr_mb/wr_s,
wr_ops/wr_s,
wr_ios/wr_s);
printf("ASYNC %dx%dx%d reads: %.3d jobs %2.3f mb in %.1f secs at ",
klen, vlen, len, jobs, rd_mb, rd_s);
printf("%2.3f mbps, %6.0f op/s, %.0f io/s\n",
rd_mb/rd_s,
rd_ops/rd_s,
rd_ios/rd_s);
}
EA *ea_new(const char *path, uint64_t bsize, int basyncs,
uint64_t *size, uint64_t *bcount, uint64_t vlun)
{
int rc = 0;
size_t plen = 0;
uint8_t *store = NULL;
EA *ea = NULL;
chunk_id_t chkid = NULL_CHUNK_ID;
chunk_ext_arg_t ext = 0;
if (!(fetch_and_or(&cflsh_blk_lib_init,1)))
{
// We need to call cblk_init once before
// we use any other cblk_ interfaces
rc = cblk_init(NULL,0);
if (rc)
{
KV_TRC_FFDC(pAT, "cblk_init failed path %s bsize %"PRIu64" "
"size %"PRIu64" bcount %"PRIu64", errno = %d",
path, bsize, *size, *bcount, errno);
goto error_exit;
}
}
ea = am_malloc(sizeof(EA));
if (NULL == ea)
{
KV_TRC_FFDC(pAT, "Out of memory path %s bsize %"PRIu64" size %"PRIu64" "
"bcount %"PRIu64", errno = %d",
path, bsize, *size, *bcount, errno);
goto error_exit;
}
// We need to check the path parameter to see if
// we are going to use memory or a file/capi
// device (to be determined by the block layer)
if ( (NULL == path) || (strlen(path) == 0) )
{
KV_TRC(pAT, "EA_STORE_TYPE_MEMORY");
// Using memory for store
ea->st_type = EA_STORE_TYPE_MEMORY;
store = malloc(*size);
if (NULL == store)
{
errno = ENOMEM;
KV_TRC_FFDC(pAT, "Out of memory for store path %s bsize %"PRIu64" "
"size %"PRIu64" bcount %"PRIu64", errno = %d",
path, bsize, *size, *bcount, errno);
goto error_exit;
}
*bcount = ((*size) / bsize);
ea->st_memory = store;
}
else
{
KV_TRC(pAT, "EA_STORE_TYPE_FILE(%s)", path);
// Using a file. We don't care if it's an actual
// file or a CAPI device, we let block layer
// decide and we just use the chunk ID that is
// passed back from the cblk_open call.
ea->st_type = EA_STORE_TYPE_FILE;
// Check to see if we need to create the store on a
// physical or virtual LUN. Previously, in GA1,
// we keyed off the size and if it was 0, then we
// asked for the LUN to be physical. Now, the user
// can specify with a flag.
if ( vlun == 0 )
{
KV_TRC(pAT, "cblk_open PHYSICAL LUN: %s", path);
chkid = cblk_open(path, basyncs, O_RDWR, ext,
CBLK_OPN_NO_INTRP_THREADS);
if (NULL_CHUNK_ID == chkid)
{
printf("cblk_open physical lun failed\n");
KV_TRC_FFDC(pAT, "cblk_open phys lun failed path:%s bsize:%ld "
"size:%ld bcount:%ld, errno:%d",
path, bsize, *size, *bcount, errno);
goto error_exit;
}
rc = cblk_get_size(chkid, (size_t *)bcount, 0);
if ( (rc != 0) || (*bcount == 0) )
{
// An error was encountered, close the chunk
cblk_close(chkid, 0);
chkid = NULL_CHUNK_ID;
KV_TRC_FFDC(pAT, "cblk_get_size failed path %s bsize %"PRIu64" "
"size %"PRIu64" bcount %"PRIu64", errno = %d",
path, bsize, *size, *bcount, errno);
goto error_exit;
}
// Set the size to be returned
*size = *bcount * bsize;
}
else
{
KV_TRC(pAT, "cblk_open VIRTUAL LUN: %s", path);
chkid = cblk_open(path, basyncs, O_RDWR, ext,
CBLK_OPN_VIRT_LUN|CBLK_OPN_NO_INTRP_THREADS);
if (NULL_CHUNK_ID == chkid)
{
printf("cblk_open virtual lun failed\n");
KV_TRC_FFDC(pAT, "cblk_open virt lun failed path:%s bsize:%ld "
"size:%ld bcount:%ld, errno:%d",
path, bsize, *size, *bcount, errno);
goto error_exit;
}
// A specific size was passed in so we try to set the
// size of the chunk.
*bcount = *size / bsize;
rc = cblk_set_size(chkid, (size_t)*bcount, 0);
if ( rc != 0 )
{
printf("cblk_set_size failed for %ld\n", *bcount);
// An error was encountered, close the chunk
cblk_close(chkid, 0);
chkid = NULL_CHUNK_ID;
KV_TRC_FFDC(pAT, "cblk_set_size failed path %s bsize %"PRIu64" "
"size %"PRIu64" bcount %"PRIu64", errno = %d",
path, bsize, *size, *bcount, errno);
goto error_exit;
}
}
// Save off the chunk ID and the device name
ea->st_flash = chkid;
plen = strlen(path) + 1;
ea->st_device = (char *)am_malloc(plen);
if (!ea->st_device)
{
cblk_close(chkid, 0);
KV_TRC_FFDC(pAT, "MALLOC st_device failed (%s) plen=%ld errno:%d",
path, plen, errno);
goto error_exit;
}
memset(ea->st_device, 0, plen);
strncpy(ea->st_device, path, plen);
}
// Fill in the EA struct
pthread_rwlock_init(&(ea->ea_rwlock), NULL);
ea->bsize = bsize;
ea->bcount = *bcount;
ea->size = *size;
KV_TRC(pAT, "path %s bsize %"PRIu64" size %"PRIu64" bcount %"PRIu64"",
path, bsize, *size, *bcount);
goto done;
error_exit:
am_free(ea);
ea = NULL;
if (!errno) {KV_TRC_FFDC(pAT, "UNSET_ERRNO"); errno=ENOSPC;}
done:
return ea;
}
int ark_delete(ARK *ark) {
int rc = 0;
int i = 0;
_ARK *_arkp = (_ARK *)ark;
scb_t *scbp = NULL;
if (NULL == ark)
{
rc = EINVAL;
KV_TRC_FFDC(pAT, "Invalid ARK control block parameter: %d", rc);
goto ark_delete_ark_err;
}
// Wait for all active threads to exit
for (i = 0; i < _arkp->nthrds; i++)
{
scbp = &(_arkp->poolthreads[i]);
scbp->poolstate = PT_EXIT;
queue_lock(scbp->rqueue);
queue_wakeup(scbp->rqueue);
queue_unlock(scbp->rqueue);
pthread_join(scbp->pooltid, NULL);
queue_free(scbp->rqueue);
queue_free(scbp->tqueue);
queue_free(scbp->ioqueue);
pthread_mutex_destroy(&(scbp->poolmutex));
pthread_cond_destroy(&(scbp->poolcond));
KV_TRC(pAT, "thread %d joined", i);
}
if (_arkp->poolthreads) am_free(_arkp->poolthreads);
if (_arkp->pts) am_free(_arkp->pts);
for ( i = 0; i < _arkp->nasyncs ; i++ )
{
pthread_cond_destroy(&(_arkp->rcbs[i].acond));
pthread_mutex_destroy(&(_arkp->rcbs[i].alock));
}
for ( i = 0; i < _arkp->ntasks; i++ )
{
bt_delete(_arkp->tcbs[i].inb);
bt_delete(_arkp->tcbs[i].oub);
am_free(_arkp->tcbs[i].vb_orig);
}
if (_arkp->iocbs)
{
am_free(_arkp->iocbs);
}
if (_arkp->tcbs)
{
am_free(_arkp->tcbs);
}
if (_arkp->rcbs)
{
am_free(_arkp->rcbs);
}
if (_arkp->ttags)
{
tag_free(_arkp->ttags);
}
if (_arkp->rtags)
{
tag_free(_arkp->rtags);
}
if (!(_arkp->flags & ARK_KV_VIRTUAL_LUN))
{
rc = ark_persist(_arkp);
if ( rc != 0 )
{
KV_TRC_FFDC(pAT, "FFDC: ark_persist failed: %d", rc);
}
}
pthread_mutex_destroy(&_arkp->mainmutex);
(void)ea_delete(_arkp->ea);
hash_free(_arkp->ht);
bl_delete(_arkp->bl);
KV_TRC(pAT, "ark_delete done %p", _arkp);
am_free(_arkp);
ark_delete_ark_err:
KV_TRC_CLOSE(pAT);
return rc;
}
int ark_create_verbose(char *path, ARK **arkret,
uint64_t size, uint64_t bsize, uint64_t hcount,
int nthrds, int nqueue, int basyncs, uint64_t flags)
{
int rc = 0;
int p_rc = 0;
uint64_t bcount = 0;
uint64_t x = 0;
int i = 0;
int tnum = 0;
int rnum = 0;
scb_t *scbp = NULL;
KV_TRC_OPEN(pAT, "arkdb");
if (NULL == arkret)
{
KV_TRC_FFDC(pAT, "Incorrect value for ARK control block: rc=EINVAL");
rc = EINVAL;
goto ark_create_ark_err;
}
if ( (flags & (ARK_KV_PERSIST_LOAD|ARK_KV_PERSIST_STORE)) &&
(flags & ARK_KV_VIRTUAL_LUN) )
{
KV_TRC_FFDC(pAT, "Invalid persistence combination with ARK flags: %016lx",
flags);
rc = EINVAL;
goto ark_create_ark_err;
}
if (nthrds <= 0)
{
KV_TRC_FFDC(pAT, "invalid nthrds:%d", nthrds);
rc = EINVAL;
goto ark_create_ark_err;
}
_ARK *ark = am_malloc(sizeof(_ARK));
if (ark == NULL) {
rc = ENOMEM;
KV_TRC_FFDC(pAT, "Out of memory allocating ARK control structure for %ld",
sizeof(_ARK));
goto ark_create_ark_err;
}
KV_TRC(pAT, "%p path(%s) size %ld bsize %ld hcount %ld "
"nthrds %d nqueue %d basyncs %d flags:%08lx",
ark, path, size, bsize, hcount,
nthrds, nqueue, basyncs, flags);
ark->bsize = bsize;
ark->rthread = 0;
ark->persload = 0;
ark->nasyncs = ((nqueue <= 0) ? ARK_MAX_ASYNC_OPS : nqueue);
ark->basyncs = basyncs;
ark->ntasks = ARK_MAX_TASK_OPS;
ark->nthrds = ARK_VERBOSE_NTHRDS_DEF; // hardcode, perf requirement
// Create the KV storage, whether that will be memory based
// or flash
ark->ea = ea_new(path, ark->bsize, basyncs, &size, &bcount,
(flags & ARK_KV_VIRTUAL_LUN));
if (ark->ea == NULL)
{
if (!errno) {KV_TRC_FFDC(pAT, "UNSET_ERRNO"); errno=ENOMEM;}
rc = errno;
KV_TRC_FFDC(pAT, "KV storage initialization failed: rc/errno:%d", rc);
goto ark_create_ea_err;
}
// Now that the "connection" to the store has been established
// we need to check to see if data was persisted from a previous
// instantiation of the KV store.
p_rc = ark_check_persistence(ark, flags);
if (p_rc > 0)
{
// We ran into an error while trying to read from
// the store.
rc = p_rc;
KV_TRC_FFDC(pAT, "Persistence check failed: %d", rc);
goto ark_create_persist_err;
}
else if (p_rc == -1)
{
KV_TRC(pAT, "NO PERSIST LOAD FLAG");
// There was no persistence data, so we just build off
// of what was passed into the API.
ark->size = size;
ark->bcount = bcount;
ark->hcount = hcount;
ark->vlimit = ARK_VERBOSE_VLIMIT_DEF;
ark->blkbits = ARK_VERBOSE_BLKBITS_DEF;
ark->grow = ARK_VERBOSE_GROW_DEF;
ark->rthread = 0;
ark->flags = flags;
ark->astart = 0;
ark->blkused = 1;
ark->ark_exit = 0;
ark->nactive = 0;
ark->pers_stats.kv_cnt = 0;
ark->pers_stats.blk_cnt = 0;
ark->pers_stats.byte_cnt = 0;
ark->pcmd = PT_IDLE;
// Create the requests and tag control blocks and queues.
x = ark->hcount / ark->nthrds;
ark->npart = x + (ark->hcount % ark->nthrds ? 1 : 0);
// Create the hash table
ark->ht = hash_new(ark->hcount);
if (ark->ht == NULL)
{
if (!errno) {KV_TRC_FFDC(pAT, "UNSET_ERRNO"); errno=ENOMEM;}
rc = errno;
KV_TRC_FFDC(pAT, "Hash initialization failed: %d", rc);
goto ark_create_ht_err;
}
// Create the block list
ark->bl = bl_new(ark->bcount, ark->blkbits);
if (ark->bl == NULL)
{
if (!errno) {KV_TRC_FFDC(pAT, "UNSET_ERRNO"); errno=ENOMEM;}
rc = errno;
KV_TRC_FFDC(pAT, "Block list initialization failed: %d", rc);
goto ark_create_bl_err;
}
if (flags & ARK_KV_PERSIST_STORE)
{
ark_persistence_calc(ark);
if (bl_reserve(ark->bl, ark->pers_max_blocks))
{goto ark_create_bl_err;}
}
}
else
{
KV_TRC(pAT, "PERSIST: %p path(%s) size %ld bsize %ld hcount %ld "
"nthrds %d nqueue %ld basyncs %d bcount %ld blkbits %ld",
ark, path, ark->size, ark->bsize, ark->hcount,
ark->nthrds, ark->nasyncs, ark->basyncs,
ark->bcount, ark->blkbits);
}
rc = pthread_mutex_init(&ark->mainmutex,NULL);
if (rc != 0)
{
KV_TRC_FFDC(pAT, "pthread_mutex_init for main mutex failed: %d", rc);
goto ark_create_pth_mutex_err;
}
ark->rtags = tag_new(ark->nasyncs);
if ( NULL == ark->rtags )
{
rc = ENOMEM;
KV_TRC_FFDC(pAT, "Tag initialization for requests failed: %d", rc);
goto ark_create_rtag_err;
}
ark->ttags = tag_new(ark->ntasks);
if ( NULL == ark->ttags )
{
rc = ENOMEM;
KV_TRC_FFDC(pAT, "Tag initialization for tasks failed: %d", rc);
goto ark_create_ttag_err;
}
ark->rcbs = am_malloc(ark->nasyncs * sizeof(rcb_t));
if ( NULL == ark->rcbs )
{
rc = ENOMEM;
KV_TRC_FFDC(pAT, "Out of memory allocation of %"PRIu64" bytes for request control blocks", (ark->nasyncs * sizeof(rcb_t)));
goto ark_create_rcbs_err;
}
ark->tcbs = am_malloc(ark->ntasks * sizeof(tcb_t));
if ( NULL == ark->tcbs )
{
rc = ENOMEM;
KV_TRC_FFDC(pAT, "Out of memory allocation of %"PRIu64" bytes for task control blocks", (ark->ntasks * sizeof(rcb_t)));
goto ark_create_tcbs_err;
}
ark->iocbs = am_malloc(ark->ntasks * sizeof(iocb_t));
if ( NULL == ark->iocbs )
{
rc = ENOMEM;
KV_TRC_FFDC(pAT, "Out of memory allocation of %"PRIu64" bytes for io control blocks", (ark->ntasks * sizeof(iocb_t)));
goto ark_create_iocbs_err;
}
ark->poolthreads = am_malloc(ark->nthrds * sizeof(scb_t));
if ( NULL == ark->poolthreads )
{
rc = ENOMEM;
KV_TRC_FFDC(pAT, "Out of memory allocation of %"PRIu64" bytes for server thread control blocks", (ark->nthrds * sizeof(scb_t)));
goto ark_create_poolthreads_err;
}
for ( rnum = 0; rnum < ark->nasyncs ; rnum++ )
{
ark->rcbs[rnum].stat = A_NULL;
pthread_cond_init(&(ark->rcbs[rnum].acond), NULL);
pthread_mutex_init(&(ark->rcbs[rnum].alock), NULL);
}
for ( tnum = 0; tnum < ark->ntasks; tnum++ )
{
ark->tcbs[tnum].inb = bt_new(0, ark->vlimit, sizeof(uint64_t),
&(ark->tcbs[tnum].inblen),
&(ark->tcbs[tnum].inb_orig));
if (ark->tcbs[tnum].inb == NULL)
{
if (!errno) {KV_TRC_FFDC(pAT, "UNSET_ERRNO"); errno=ENOMEM;}
rc = errno;
KV_TRC_FFDC(pAT, "Bucket allocation for inbuffer failed: %d", rc);
goto ark_create_taskloop_err;
}
ark->tcbs[tnum].oub = bt_new(0, ark->vlimit, sizeof(uint64_t),
&(ark->tcbs[tnum].oublen),
&(ark->tcbs[tnum].oub_orig));
if (ark->tcbs[tnum].oub == NULL)
{
if (!errno) {KV_TRC_FFDC(pAT, "UNSET_ERRNO"); errno=ENOMEM;}
rc = errno;
KV_TRC_FFDC(pAT, "Bucket allocation for outbuffer failed: %d", rc);
goto ark_create_taskloop_err;
}
//ark->tcbs[tnum].vbsize = bsize * 1024;
ark->tcbs[tnum].vbsize = bsize * 256;
ark->tcbs[tnum].vb_orig = am_malloc(ark->tcbs[tnum].vbsize);
if (ark->tcbs[tnum].vb_orig == NULL)
{
rc = ENOMEM;
KV_TRC_FFDC(pAT, "Out of memory allocation for %"PRIu64" bytes for variable size buffer", (bsize * 1024));
goto ark_create_taskloop_err;
}
ark->tcbs[tnum].vb = ptr_align(ark->tcbs[tnum].vb_orig);
}
*arkret = (void *)ark;
ark->pts = (PT *)am_malloc(sizeof(PT) * ark->nthrds);
if ( ark->pts == NULL )
{
rc = ENOMEM;
KV_TRC_FFDC(pAT, "Out of memory allocation for %"PRIu64" bytes for server thread data", (sizeof(PT) * ark->nthrds));
goto ark_create_taskloop_err;
}
for (i = 0; i < ark->nthrds; i++) {
PT *pt = &(ark->pts[i]);
scbp = &(ark->poolthreads[i]);
memset(scbp, 0, sizeof(scb_t));
// Start off the random start point for this thread
// at -1, to show that it has not been part of a
// ark_random call.
scbp->rlast = -1;
scbp->holds = 0;
scbp->poolstate = PT_RUN;
scbp->poolstats.io_cnt = 0;
scbp->poolstats.ops_cnt = 0;
scbp->poolstats.kv_cnt = 0;
scbp->poolstats.blk_cnt = 0;
scbp->poolstats.byte_cnt = 0;
pthread_mutex_init(&(scbp->poolmutex), NULL);
pthread_cond_init(&(scbp->poolcond), NULL);
scbp->rqueue = queue_new(ark->nasyncs);
scbp->tqueue = queue_new(ark->ntasks);
scbp->ioqueue = queue_new(ark->ntasks);
pt->id = i;
pt->ark = ark;
rc = pthread_create(&(scbp->pooltid), NULL, pool_function, pt);
if (rc != 0)
{
KV_TRC_FFDC(pAT, "pthread_create of server thread failed: %d", rc);
goto ark_create_poolloop_err;
}
}
#if 0
while (ark->nactive < ark->nthrds) {
usleep(1);
//printf("Create waiting %d/%d\n", ark->nactive, ark->nthrds);
}
#endif
ark->pcmd = PT_RUN;
goto ark_create_return;
ark_create_poolloop_err:
for (; i >= 0; i--)
{
scbp = &(ark->poolthreads[i]);
if (scbp->pooltid != 0)
{
queue_lock(scbp->rqueue);
queue_wakeup(scbp->rqueue);
queue_unlock(scbp->rqueue);
pthread_join(scbp->pooltid, NULL);
pthread_mutex_destroy(&(scbp->poolmutex));
pthread_cond_destroy(&(scbp->poolcond));
if ( scbp->rqueue != NULL )
{
queue_free(scbp->rqueue);
}
if ( scbp->tqueue != NULL )
{
queue_free(scbp->tqueue);
}
if ( scbp->ioqueue != NULL )
{
queue_free(scbp->ioqueue);
}
}
}
if ( ark->pts != NULL )
{
am_free(ark->pts);
}
ark_create_taskloop_err:
for ( tnum = 0; tnum < ark->ntasks; tnum++ )
{
if (ark->tcbs[tnum].inb)
{
bt_delete(ark->tcbs[tnum].inb);
}
if (ark->tcbs[tnum].oub)
{
bt_delete(ark->tcbs[tnum].oub);
}
if (ark->tcbs[tnum].vb_orig)
{
am_free(ark->tcbs[tnum].vb_orig);
}
}
for (rnum = 0; rnum < ark->nasyncs; rnum++)
{
pthread_cond_destroy(&(ark->rcbs[rnum].acond));
pthread_mutex_destroy(&(ark->rcbs[rnum].alock));
}
if ( ark->poolthreads != NULL )
{
am_free(ark->poolthreads);
}
ark_create_poolthreads_err:
if (ark->iocbs)
{
am_free(ark->iocbs);
}
ark_create_iocbs_err:
if (ark->tcbs)
{
am_free(ark->tcbs);
}
ark_create_tcbs_err:
if (ark->rcbs)
{
am_free(ark->rcbs);
}
ark_create_rcbs_err:
if (ark->ttags)
{
tag_free(ark->ttags);
}
ark_create_ttag_err:
if (ark->rtags)
{
tag_free(ark->rtags);
}
ark_create_rtag_err:
pthread_mutex_destroy(&ark->mainmutex);
ark_create_pth_mutex_err:
bl_delete(ark->bl);
ark_create_bl_err:
hash_free(ark->ht);
ark_create_ht_err:
ark_create_persist_err:
ea_delete(ark->ea);
ark_create_ea_err:
am_free(ark);
*arkret = NULL;
ark_create_ark_err:
KV_TRC_CLOSE(pAT);
ark_create_return:
return rc;
}
int ark_check_persistence(_ARK *_arkp, uint64_t flags)
{
int32_t rc = -1;
char *p_data_orig = NULL;
char *p_data = NULL;
ark_io_list_t *bl_array = NULL;
p_cntr_t *pptr = NULL;
P_ARK_t *pcfg = NULL;
hash_t *htp = NULL;
BL *blp = NULL;
uint64_t rdblks = 0;
if (flags & ARK_KV_PERSIST_LOAD) {KV_TRC(pAT, "PERSIST_LOAD");}
// Ignore the persistence data and load from scratch
if ( (!(flags & ARK_KV_PERSIST_LOAD)) || (flags & ARK_KV_VIRTUAL_LUN) )
{
return -1;
}
p_data_orig = am_malloc(_arkp->bsize);
if (p_data_orig == NULL)
{
KV_TRC_FFDC(pAT, "Out of memory allocating %"PRIu64" bytes for the first "
"persistence block", _arkp->bsize);
rc = ENOMEM;
}
else
{
p_data = ptr_align(p_data_orig);
bl_array = bl_chain_no_bl(0, 1);
rc = ea_async_io(_arkp->ea, ARK_EA_READ, (void *)p_data,
bl_array, 1, 1);
am_free(bl_array);
}
if (rc == 0)
{
// We've read the first block. We check to see if
// persistence data is present and if so, then
// read the rest of the data from the flash.
pptr = (p_cntr_t *)p_data;
_arkp->persdata = p_data_orig;
if ( memcmp(pptr->p_cntr_magic, ARK_P_MAGIC,
sizeof(pptr->p_cntr_magic) != 0))
{
KV_TRC_FFDC(pAT, "No magic number found in persistence data: %d", EINVAL);
// The magic number does not match so data is either
// not present or is corrupted.
rc = -1;
}
else
{
// Now we check version and the first persistence data
// needs to be the ARK_PERSIST_CONFIG block
if (pptr->p_cntr_version != ARK_P_VERSION_1 &&
pptr->p_cntr_version != ARK_P_VERSION_2)
{
KV_TRC_FFDC(pAT, "Invalid / unsupported version: %"PRIu64"",
pptr->p_cntr_version);
rc = EINVAL;
}
else
{
// Read in the rest of the persistence data
pcfg = (P_ARK_t *)(pptr->p_cntr_data + pptr->p_cntr_cfg_offset);
rdblks = pcfg->pblocks;
if (rdblks > 1)
{
p_data_orig = am_realloc(p_data_orig, (rdblks * _arkp->bsize));
if (p_data_orig == NULL)
{
KV_TRC_FFDC(pAT, "Out of memory allocating %"PRIu64" bytes for "
"full persistence block",
(rdblks * _arkp->bsize));
rc = ENOMEM;
}
else
{
p_data = ptr_align(p_data_orig);
bl_array = bl_chain_no_bl(0, rdblks);
if (bl_array == NULL)
{
KV_TRC_FFDC(pAT, "Out of memory allocating %"PRIu64" blocks for "
"full persistence data", rdblks);
rc = ENOMEM;
}
}
// We are still good to read the rest of the data
// from the flash
if (rc == 0)
{
KV_TRC(pAT, "PERSIST_RD rdblks:%ld", rdblks);
rc = ea_async_io(_arkp->ea, ARK_EA_READ, (void *)p_data,
bl_array, rdblks, 1);
am_free(bl_array);
pptr = (p_cntr_t *)p_data;
pcfg = (P_ARK_t *)(pptr->p_cntr_data + pptr->p_cntr_cfg_offset);
_arkp->persdata = p_data_orig;
}
}
}
}
}
// If rc == 0, that means we have persistence data
if (rc == 0)
{
KV_TRC(pAT, "PERSIST_META size %ld bsize %ld hcount %ld bcount %ld "
"nthrds %d nasyncs %d basyncs %d blkbits %ld version:%ld",
pcfg->size, pcfg->bsize, pcfg->hcount, pcfg->bcount,
pcfg->nthrds, pcfg->nasyncs, pcfg->basyncs, pcfg->blkbits,
pptr->p_cntr_version);
_arkp->persload = 1;
_arkp->size = pcfg->size;
_arkp->flags = flags;
_arkp->bsize = pcfg->bsize;
_arkp->bcount = pcfg->bcount;
_arkp->blkbits = pcfg->blkbits;
_arkp->grow = pcfg->grow;
_arkp->hcount = pcfg->hcount;
_arkp->vlimit = pcfg->vlimit;
_arkp->blkused = pcfg->blkused;
_arkp->pers_stats.kv_cnt = pcfg->pstats.kv_cnt;
_arkp->pers_stats.blk_cnt = pcfg->pstats.blk_cnt;
_arkp->pers_stats.byte_cnt = pcfg->pstats.byte_cnt;
KV_TRC(pAT, "ARK_META size %ld bsize %ld hcount %ld bcount %ld "
"nthrds %d nasyncs %ld basyncs %d blkbits %ld",
_arkp->size, _arkp->bsize, _arkp->hcount, _arkp->bcount,
_arkp->nthrds, _arkp->nasyncs, _arkp->basyncs, _arkp->blkbits);
htp = (hash_t *)(pptr->p_cntr_data + pptr->p_cntr_ht_offset);
_arkp->ht = hash_new(htp->n);
if (_arkp->ht == NULL)
{
if (!errno) {KV_TRC_FFDC(pAT, "UNSET_ERRNO"); errno=ENOMEM;}
rc = errno;
KV_TRC_FFDC(pAT, "ht_new failed: n:%ld rc:%d", htp->n, rc);
goto error_exit;
}
memcpy(_arkp->ht, htp, pptr->p_cntr_ht_size);
blp = (BL *)(pptr->p_cntr_data + pptr->p_cntr_bl_offset);
_arkp->bl = bl_new(blp->n, blp->w);
if (_arkp->bl == NULL)
{
if (!errno) {KV_TRC_FFDC(pAT, "UNSET_ERRNO"); errno=ENOMEM;}
rc = errno;
KV_TRC_FFDC(pAT, "bl_new failed: n:%ld w:%ld rc:%d",
blp->n, blp->w, rc);
goto error_exit;
}
_arkp->bl->count = blp->count;
_arkp->bl->head = blp->head;
_arkp->bl->hold = blp->hold;
_arkp->bl->top = blp->top;
if (pptr->p_cntr_version == ARK_P_VERSION_1)
{
IV *piv = (IV *)(pptr->p_cntr_data + pptr->p_cntr_bliv_offset);
KV_TRC(pAT, "PERSIST_VERSION_1 LOADED");
_arkp->bl->top = _arkp->bl->n;
// copy IV->data from piv->data
memcpy(_arkp->bl->list->data,
piv->data,
pptr->p_cntr_bliv_size);
}
else if (pptr->p_cntr_version == ARK_P_VERSION_2)
{
KV_TRC(pAT, "PERSIST_VERSION_2 LOADED");
// copy IV->data from bliv_offset
memcpy(_arkp->bl->list->data,
pptr->p_cntr_data + pptr->p_cntr_bliv_offset,
pptr->p_cntr_bliv_size);
}
else
{
rc = EINVAL;
KV_TRC_FFDC(pAT, "bad persistent version number: ver:%ld",
pptr->p_cntr_version);
goto error_exit;
}
KV_TRC(pAT, "BL_META: n:%ld count:%ld head:%ld hold:%ld top:%ld",
_arkp->bl->n, _arkp->bl->count, _arkp->bl->head, _arkp->bl->hold,
_arkp->bl->top);
}
error_exit:
am_free(p_data_orig);
return rc;
}
int ark_persist(_ARK *_arkp)
{
int32_t rc = 0;
uint64_t tot_bytes = 0;
uint64_t wrblks = 0;
char *p_data_orig = NULL;
char *p_data = NULL;
p_cntr_t *pptr = NULL;
char *dptr = NULL;
P_ARK_t *pcfg = NULL;
ark_io_list_t *bl_array = NULL;
if ( (_arkp->ea->st_type == EA_STORE_TYPE_MEMORY) ||
!(_arkp->flags & ARK_KV_PERSIST_STORE) )
{
return 0;
}
ark_persistence_calc(_arkp);
// allocate write buffer
tot_bytes = _arkp->pers_max_blocks * _arkp->bsize;
p_data_orig = am_malloc(tot_bytes);
if (p_data_orig == NULL)
{
KV_TRC_FFDC(pAT, "Out of memory allocating %"PRIu64" bytes for "
"persistence data", tot_bytes);
return ENOMEM;
}
memset(p_data_orig, 0, tot_bytes);
p_data = ptr_align(p_data_orig);
// Record cntr data
pptr = (p_cntr_t *)p_data;
memcpy(pptr->p_cntr_magic, ARK_P_MAGIC, sizeof(pptr->p_cntr_magic));
pptr->p_cntr_version = ARK_P_VERSION_2;
pptr->p_cntr_size = sizeof(p_cntr_t);
// Record configuration info
pcfg = (P_ARK_t*)pptr->p_cntr_data;
pcfg->flags = _arkp->flags;
pcfg->size = _arkp->ea->size;
pcfg->bsize = _arkp->bsize;
pcfg->bcount = _arkp->bcount;
pcfg->blkbits = _arkp->blkbits;
pcfg->grow = _arkp->blkbits;
pcfg->hcount = _arkp->hcount;
pcfg->vlimit = _arkp->vlimit;
pcfg->blkused = _arkp->blkused;
pcfg->nasyncs = _arkp->nasyncs;
pcfg->basyncs = _arkp->basyncs;
pcfg->ntasks = _arkp->ntasks;
pcfg->nthrds = _arkp->nthrds;
ark_persist_stats(_arkp, &(pcfg->pstats));
pptr->p_cntr_cfg_offset = 0;
pptr->p_cntr_cfg_size = sizeof(P_ARK_t);
dptr = pptr->p_cntr_data;
// Record hash info
dptr += pptr->p_cntr_cfg_size;
pptr->p_cntr_ht_offset = dptr - pptr->p_cntr_data;
pptr->p_cntr_ht_size = sizeof(hash_t) + (_arkp->ht->n * sizeof(uint64_t));
memcpy(dptr, _arkp->ht, pptr->p_cntr_ht_size);
// Record block list info
dptr += pptr->p_cntr_ht_size;
pptr->p_cntr_bl_offset = dptr - pptr->p_cntr_data;
pptr->p_cntr_bl_size = sizeof(BL);
memcpy(dptr, _arkp->bl, pptr->p_cntr_bl_size);
// Record IV list info
dptr += pptr->p_cntr_bl_size;
pptr->p_cntr_bliv_offset = dptr - pptr->p_cntr_data;
// bliv_size = bytes in bl->list->data[cs_blocks + kvdata_blocks]
// add 2 to top because of how IV->data chaining works
pptr->p_cntr_bliv_size = divup((_arkp->bl->top+2) * _arkp->bl->w, 8);
memcpy(dptr, _arkp->bl->list->data, pptr->p_cntr_bliv_size);
// Calculate wrblks: number of persist metadata blocks to write
tot_bytes = _arkp->pers_cs_bytes + pptr->p_cntr_bliv_size;
wrblks = pcfg->pblocks = divup(tot_bytes, _arkp->bsize);
KV_TRC(pAT, "PERSIST_WR dev:%s top:%ld wrblks:%ld vs pers_max_blocks:%ld",
_arkp->ea->st_device, _arkp->bl->top, pcfg->pblocks,
_arkp->pers_max_blocks);
bl_array = bl_chain_blocks(_arkp->bl, 0, wrblks);
if ( NULL == bl_array )
{
KV_TRC_FFDC(pAT, "Out of memory allocating %"PRIu64" blocks for block list",
wrblks);
rc = ENOMEM;
}
else
{
rc = ea_async_io(_arkp->ea, ARK_EA_WRITE, (void *)p_data,
bl_array, wrblks, _arkp->nthrds);
am_free(bl_array);
}
KV_TRC(pAT, "PERSIST_DATA_STORED rc:%d", rc);
am_free(p_data_orig);
return rc;
}
