/*
* Function: Open storage record for passed in udf record
* also set up flag like exists / read et al.
*
* Parameters:
* urec : UDF record
*
* Return value : 0 on success
* -1 if the record's bin count exceeds the UDF limit
*
* Callers:
* udf_record_open
*
* Note: There are no checks, so the caller has to make sure that all
* protections are taken and all checks are done.
*
* Side effect:
* Counters will be reset
* flag will be set
* bins will be opened
*/
int
udf_storage_record_open(udf_record *urecord)
{
cf_debug_digest(AS_UDF, &urecord->tr->keyd, "[ENTER] Opening record key:");
as_storage_rd *rd = urecord->rd;
as_index *r = urecord->r_ref->r;
as_transaction *tr = urecord->tr;
int rv = as_storage_record_open(tr->rsv.ns, r, rd, &r->key);
if (0 != rv) {
cf_warning(AS_UDF, "Could not open record !! %d", rv);
return rv;
}
rd->n_bins = as_bin_get_n_bins(r, rd);
if (rd->n_bins > UDF_RECORD_BIN_ULIMIT) {
cf_warning(AS_UDF, "record has too many bins (%d) for UDF processing", rd->n_bins);
as_storage_record_close(r, rd);
return -1;
}
// if multibin storage, we will use urecord->stack_bins, so set the size appropriately
if ( ! tr->rsv.ns->storage_data_in_memory && ! tr->rsv.ns->single_bin ) {
rd->n_bins = sizeof(urecord->stack_bins) / sizeof(as_bin);
}
rd->bins = as_bin_get_all(r, rd, urecord->stack_bins);
urecord->starting_memory_bytes = as_storage_record_get_n_bytes_memory(rd);
as_storage_record_get_key(rd);
urecord->flag |= UDF_RECORD_FLAG_STORAGE_OPEN;
if (urecord->flag & UDF_RECORD_FLAG_IS_SUBRECORD) {
urecord->lrecord->subrec_io++;
}
cf_detail_digest(AS_UDF, &tr->keyd, "Storage Open: Rec(%p) flag(%x) Digest:", urecord, urecord->flag );
if (urecord->flag & UDF_RECORD_FLAG_IS_SUBRECORD) {
as_ldt_subrec_storage_validate(rd, "Reading");
}
return 0;
}
/**
* aerospike::create(record)
* Function: udf_aerospike_rec_create
*
* Parameters:
* as - as_aerospike
* rec - as_rec
*
* Return Values:
* 1 if record is being read or on a create, it already exists
* o/w return value of udf_aerospike__execute_updates
*
* Description:
* Create a new record in local storage.
* The record will only be created if it does not exist.
* This assumes the record has a digest that is valid for local storage.
*
* Synchronization : object lock acquired by the transaction thread executing UDF.
* Partition reservation takes place just before the transaction starts executing
* ( look for as_partition_reserve_udf in thr_tsvc.c )
*
* Callers:
* lua interfacing function, mod_lua_aerospike_rec_create
* The return value of udf_aerospike_rec_create is pushed on to the lua stack
*
* Notes:
* The 'read' and 'exists' flag of udf_record are set to true.
*/
static int
udf_aerospike_rec_create(const as_aerospike * as, const as_rec * rec)
{
int ret = udf_aerospike_param_check(as, rec, __FILE__, __LINE__);
if (ret) {
return ret;
}
udf_record * urecord = (udf_record *) as_rec_source(rec);
// make sure record isn't already successfully read
if (urecord->flag & UDF_RECORD_FLAG_OPEN) {
cf_detail(AS_UDF, "udf_aerospike_rec_create: Record Already Exists");
return 1;
}
as_transaction *tr = urecord->tr;
as_index_ref *r_ref = urecord->r_ref;
as_storage_rd *rd = urecord->rd;
as_index_tree *tree = tr->rsv.tree;
if (urecord->flag & UDF_RECORD_FLAG_IS_SUBRECORD) {
tree = tr->rsv.sub_tree;
}
// make sure we got the record as a create
int rv = as_record_get_create(tree, &tr->keyd, r_ref, tr->rsv.ns);
cf_detail_digest(AS_UDF, &tr->keyd, "Creating %sRecord",
(urecord->flag & UDF_RECORD_FLAG_IS_SUBRECORD) ? "Sub" : "");
// rv 0 means record exists, 1 means create, < 0 means fail
// TODO: Verify correct result codes.
if (rv == 0) {
cf_warning(AS_UDF, "udf_aerospike_rec_create: Record Already Exists 2");
as_record_done(r_ref, tr->rsv.ns);
bzero(r_ref, sizeof(as_index_ref));
return 1;
} else if (rv < 0) {
cf_warning(AS_UDF, "udf_aerospike_rec_create: Record Open Failed with rv=%d", rv);
return rv;
}
// Associates the set name with the storage rec and index
if(tr->msgp) {
// Set the set name to index and close record if the setting the set name
// is not successful
int rv_set = as_record_set_set_from_msg(r_ref->r, tr->rsv.ns, &tr->msgp->msg);
if (rv_set != 0) {
cf_warning(AS_UDF, "udf_aerospike_rec_create: Failed to set setname");
as_record_done(r_ref, tr->rsv.ns);
// TODO bzero is expensive. Switch to use flag.
bzero(r_ref, sizeof(as_index_ref));
return 4;
}
}
urecord->flag |= UDF_RECORD_FLAG_OPEN;
cf_detail(AS_UDF, "Open %p %x %"PRIx64"", urecord, urecord->flag, *(uint64_t *)&tr->keyd);
as_index *r = r_ref->r;
// open up storage
as_storage_record_create(urecord->tr->rsv.ns, urecord->r_ref->r,
urecord->rd, &urecord->tr->keyd);
cf_detail(AS_UDF, "as_storage_record_create: udf_aerospike_rec_create: r %p rd %p",
urecord->r_ref->r, urecord->rd);
// if multibin storage, we will use urecord->stack_bins, so set the size appropriately
if ( ! rd->ns->storage_data_in_memory && ! rd->ns->single_bin ) {
rd->n_bins = sizeof(urecord->stack_bins) / sizeof(as_bin);
}
// side effect: will set the unused bins to properly unused
rd->bins = as_bin_get_all(r, rd, urecord->stack_bins);
urecord->flag |= UDF_RECORD_FLAG_STORAGE_OPEN;
// If the message has a key, apply it to the record.
as_msg_field* f = as_msg_field_get(&tr->msgp->msg, AS_MSG_FIELD_TYPE_KEY);
if (f) {
rd->key_size = as_msg_field_get_value_sz(f);
rd->key = f->data;
}
cf_detail(AS_UDF, "Storage Open %p %x %"PRIx64"", urecord, urecord->flag, *(uint64_t *)&tr->keyd);
cf_detail(AS_UDF, "udf_aerospike_rec_create: Record created %d", urecord->flag);
int rc = udf_aerospike__execute_updates(urecord);
if(rc) {
// Creating the udf record failed, destroy the as_record
if (!as_bin_inuse_has(urecord->rd)) {
udf_aerospike_rec_remove(as, rec);
}
}
return rc;
}
/**
* aerospike::create(record)
* Function: udf_aerospike_rec_create
*
* Parameters:
* as - as_aerospike
* rec - as_rec
*
* Return Values:
* 1 if record is being read or on a create, it already exists
* o/w return value of udf_aerospike__execute_updates
*
* Description:
* Create a new record in local storage.
* The record will only be created if it does not exist.
* This assumes the record has a digest that is valid for local storage.
*
* Synchronization : object lock acquired by the transaction thread executing UDF.
* Partition reservation takes place just before the transaction starts executing
* ( look for as_partition_reserve_udf in thr_tsvc.c )
*
* Callers:
* lua interfacing function, mod_lua_aerospike_rec_create
* The return value of udf_aerospike_rec_create is pushed on to the lua stack
*
* Notes:
* The 'read' and 'exists' flag of udf_record are set to true.
*/
static int
udf_aerospike_rec_create(const as_aerospike * as, const as_rec * rec)
{
int ret = udf_aerospike_param_check(as, rec, __FILE__, __LINE__);
if (ret) {
return ret;
}
udf_record * urecord = (udf_record *) as_rec_source(rec);
// make sure record isn't already successfully read
if (urecord->flag & UDF_RECORD_FLAG_OPEN) {
cf_detail(AS_UDF, "udf_aerospike_rec_create: Record Already Exists");
return 1;
}
as_transaction *tr = urecord->tr;
as_index_ref *r_ref = urecord->r_ref;
as_storage_rd *rd = urecord->rd;
as_index_tree *tree = tr->rsv.tree;
bool is_subrec = false;
if (urecord->flag & UDF_RECORD_FLAG_IS_SUBRECORD) {
tree = tr->rsv.sub_tree;
is_subrec = true;
}
// make sure we got the record as a create
bool is_create = false;
int rv = as_record_get_create(tree, &tr->keyd, r_ref, tr->rsv.ns, is_subrec);
cf_detail_digest(AS_UDF, &tr->keyd, "Creating %sRecord",
(urecord->flag & UDF_RECORD_FLAG_IS_SUBRECORD) ? "Sub" : "");
// rv 0 means record exists, 1 means create, < 0 means fail
// TODO: Verify correct result codes.
if (rv == 1) {
is_create = true;
} else if (rv == 0) {
// If it's an expired record, pretend it's a fresh create.
if (as_record_is_expired(r_ref->r)) {
as_record_destroy(r_ref->r, tr->rsv.ns);
as_record_initialize(r_ref, tr->rsv.ns);
cf_atomic_int_incr(&tr->rsv.ns->n_objects);
is_create = true;
} else {
cf_warning(AS_UDF, "udf_aerospike_rec_create: Record Already Exists 2");
as_record_done(r_ref, tr->rsv.ns);
// DO NOT change it has special meaning for caller
return 1;
}
} else if (rv < 0) {
cf_warning(AS_UDF, "udf_aerospike_rec_create: Record Open Failed with rv=%d", rv);
return rv;
}
// Associates the set name with the storage rec and index
if (tr->msgp) {
// Set the set name to index and close record if the setting the set name
// is not successful
int rv_set = as_transaction_has_set(tr) ?
as_record_set_set_from_msg(r_ref->r, tr->rsv.ns, &tr->msgp->msg) : 0;
if (rv_set != 0) {
cf_warning(AS_UDF, "udf_aerospike_rec_create: Failed to set setname");
if (is_create) {
as_index_delete(tree, &tr->keyd);
}
as_record_done(r_ref, tr->rsv.ns);
return 4;
}
}
urecord->flag |= UDF_RECORD_FLAG_OPEN;
cf_detail(AS_UDF, "Open %p %x %"PRIx64"", urecord, urecord->flag, *(uint64_t *)&tr->keyd);
as_index *r = r_ref->r;
// open up storage
as_storage_record_create(urecord->tr->rsv.ns, urecord->r_ref->r,
urecord->rd, &urecord->tr->keyd);
cf_detail(AS_UDF, "as_storage_record_create: udf_aerospike_rec_create: r %p rd %p",
urecord->r_ref->r, urecord->rd);
// If the message has a key, apply it to the record.
if (! get_msg_key(tr, rd)) {
cf_warning(AS_UDF, "udf_aerospike_rec_create: Can't store key");
if (is_create) {
as_index_delete(tree, &tr->keyd);
}
as_record_done(r_ref, tr->rsv.ns);
urecord->flag &= ~UDF_RECORD_FLAG_OPEN;
return 4;
}
// if multibin storage, we will use urecord->stack_bins, so set the size appropriately
if ( ! rd->ns->storage_data_in_memory && ! rd->ns->single_bin ) {
rd->n_bins = sizeof(urecord->stack_bins) / sizeof(as_bin);
}
// side effect: will set the unused bins to properly unused
rd->bins = as_bin_get_all(r, rd, urecord->stack_bins);
urecord->flag |= UDF_RECORD_FLAG_STORAGE_OPEN;
cf_detail(AS_UDF, "Storage Open %p %x %"PRIx64"", urecord, urecord->flag, *(uint64_t *)&tr->keyd);
cf_detail(AS_UDF, "udf_aerospike_rec_create: Record created %d", urecord->flag);
int rc = udf_aerospike__execute_updates(urecord);
if (rc) {
// Creating the udf record failed, destroy the as_record
cf_warning(AS_UDF, "udf_aerospike_rec_create: failure executing record updates (%d)", rc);
if (!as_bin_inuse_has(urecord->rd)) {
udf_aerospike_rec_remove(as, rec);
}
}
return rc;
}
int
write_replica(as_partition_reservation* rsv, cf_digest* keyd,
uint8_t* pickled_buf, size_t pickled_sz,
const as_rec_props* p_rec_props, as_generation generation,
uint32_t void_time, uint64_t last_update_time, cf_node master,
uint32_t info, ldt_prole_info* linfo)
{
as_namespace* ns = rsv->ns;
if (! as_storage_has_space(rsv->ns)) {
cf_warning(AS_RW, "{%s} write_replica: drives full", ns->name);
return AS_PROTO_RESULT_FAIL_PARTITION_OUT_OF_SPACE;
}
as_index_tree* tree = rsv->tree;
bool is_subrec = false;
bool is_ldt_parent = false;
if (ns->ldt_enabled) {
if ((info & RW_INFO_LDT_SUBREC) != 0 || (info & RW_INFO_LDT_ESR) != 0) {
tree = rsv->sub_tree;
is_subrec = true;
}
else if ((info & RW_INFO_LDT_PARENTREC) != 0) {
is_ldt_parent = true;
}
}
as_index_ref r_ref;
r_ref.skip_lock = false;
int rv = as_record_get_create(tree, keyd, &r_ref, ns, is_subrec);
if (rv < 0) {
cf_warning_digest(AS_RW, keyd, "{%s} write_replica: fail as_record_get_create() ", ns->name);
return AS_PROTO_RESULT_FAIL_UNKNOWN;
}
as_record* r = r_ref.r;
as_storage_rd rd;
bool is_create = false;
if (rv == 1) {
as_storage_record_create(ns, r, &rd, keyd);
is_create = true;
}
else {
as_storage_record_open(ns, r, &rd, keyd);
}
bool has_sindex = (info & RW_INFO_SINDEX_TOUCHED) != 0;
rd.ignore_record_on_device = ! has_sindex && ! is_ldt_parent;
rd.n_bins = as_bin_get_n_bins(r, &rd);
// TODO - we really need an inline utility for this!
uint16_t newbins = ntohs(*(uint16_t*)pickled_buf);
if (! rd.ns->storage_data_in_memory && ! rd.ns->single_bin &&
newbins > rd.n_bins) {
rd.n_bins = newbins;
}
as_bin stack_bins[rd.ns->storage_data_in_memory ? 0 : rd.n_bins];
rd.bins = as_bin_get_all(r, &rd, stack_bins);
uint32_t stack_particles_sz = rd.ns->storage_data_in_memory ?
0 : as_record_buf_get_stack_particles_sz(pickled_buf);
uint8_t stack_particles[stack_particles_sz + 256];
uint8_t* p_stack_particles = stack_particles;
// + 256 for LDT control bin, to hold version.
if (! ldt_get_prole_version(rsv, keyd, linfo, info, &rd, is_create)) {
if (is_create) {
as_index_delete(tree, keyd);
}
as_storage_record_close(r, &rd);
as_record_done(&r_ref, ns);
return AS_PROTO_RESULT_FAIL_UNKNOWN;
}
uint64_t memory_bytes = 0;
if (! is_create) {
memory_bytes = as_storage_record_get_n_bytes_memory(&rd);
}
as_record_set_properties(&rd, p_rec_props);
if (as_record_unpickle_replace(r, &rd, pickled_buf, pickled_sz,
&p_stack_particles, has_sindex) != 0) {
if (is_create) {
as_index_delete(tree, keyd);
}
as_storage_record_close(r, &rd);
as_record_done(&r_ref, ns);
return AS_PROTO_RESULT_FAIL_UNKNOWN; // TODO - better granularity?
}
r->generation = generation;
r->void_time = void_time;
r->last_update_time = last_update_time;
as_storage_record_adjust_mem_stats(&rd, memory_bytes);
uint64_t version_to_set = 0;
bool set_version = false;
if (is_ldt_parent) {
if (linfo->replication_partition_version_match &&
linfo->ldt_prole_version_set) {
version_to_set = linfo->ldt_prole_version;
set_version = true;
}
else if (! linfo->replication_partition_version_match) {
version_to_set = linfo->ldt_source_version;
set_version = true;
}
}
if (set_version) {
int ldt_rv = as_ldt_parent_storage_set_version(&rd, version_to_set,
p_stack_particles, __FILE__, __LINE__);
if (ldt_rv < 0) {
cf_warning(AS_LDT, "write_replica: LDT parent storage version set failed %d", ldt_rv);
// TODO - roll back.
}
}
bool is_delete = false;
if (! as_bin_inuse_has(&rd)) {
// A master write that deletes a record by deleting (all) bins sends a
// binless pickle that ends up here.
is_delete = true;
as_index_delete(tree, keyd);
}
as_storage_record_write(r, &rd);
as_storage_record_close(r, &rd);
uint16_t set_id = as_index_get_set_id(r);
as_record_done(&r_ref, ns);
// Don't send an XDR delete if it's disallowed.
if (is_delete && ! is_xdr_delete_shipping_enabled()) {
// TODO - should we also not ship if there was no record here before?
return AS_PROTO_RESULT_OK;
}
// Do XDR write if the write is a non-XDR write or forwarding is enabled.
if ((info & RW_INFO_XDR) == 0 ||
is_xdr_forwarding_enabled() || ns->ns_forward_xdr_writes) {
xdr_write(ns, *keyd, generation, master, is_delete, set_id, NULL);
}
return AS_PROTO_RESULT_OK;
}
// Build response to batch request.
static void
batch_build_response(batch_transaction* btr, cf_buf_builder** bb_r)
{
as_namespace* ns = btr->ns;
batch_digests *bmds = btr->digests;
bool get_data = btr->get_data;
uint32_t yield_count = 0;
for (int i = 0; i < bmds->n_digests; i++)
{
batch_digest *bmd = &bmds->digest[i];
if (bmd->done == false) {
// try to get the key
as_partition_reservation rsv;
AS_PARTITION_RESERVATION_INIT(rsv);
cf_node other_node = 0;
uint64_t cluster_key;
if (! *bb_r) {
*bb_r = cf_buf_builder_create_size(1024 * 4);
}
int rv = as_partition_reserve_read(ns, as_partition_getid(bmd->keyd), &rsv, &other_node, &cluster_key);
if (rv == 0) {
cf_atomic_int_incr(&g_config.batch_tree_count);
as_index_ref r_ref;
r_ref.skip_lock = false;
int rec_rv = as_record_get(rsv.tree, &bmd->keyd, &r_ref, ns);
if (rec_rv == 0) {
as_index *r = r_ref.r;
// Check to see this isn't an expired record waiting to die.
if (r->void_time && r->void_time < as_record_void_time_get()) {
as_msg_make_error_response_bufbuilder(&bmd->keyd, AS_PROTO_RESULT_FAIL_NOTFOUND, bb_r, ns->name);
}
else {
// Make sure it's brought in from storage if necessary.
as_storage_rd rd;
if (get_data) {
as_storage_record_open(ns, r, &rd, &r->key);
rd.n_bins = as_bin_get_n_bins(r, &rd);
}
// Note: this array must stay in scope until the
// response for this record has been built, since in the
// get data w/ record on device case, it's copied by
// reference directly into the record descriptor.
as_bin stack_bins[!get_data || rd.ns->storage_data_in_memory ? 0 : rd.n_bins];
if (get_data) {
// Figure out which bins you want - for now, all.
rd.bins = as_bin_get_all(r, &rd, stack_bins);
rd.n_bins = as_bin_inuse_count(&rd);
}
as_msg_make_response_bufbuilder(r, (get_data ? &rd : NULL), bb_r, !get_data, (get_data ? NULL : ns->name), true, false, btr->binlist);
if (get_data) {
as_storage_record_close(r, &rd);
}
}
as_record_done(&r_ref, ns);
}
else {
// TODO - what about empty records?
cf_debug(AS_BATCH, "batch_build_response: as_record_get returned %d : key %"PRIx64, rec_rv, *(uint64_t *)&bmd->keyd);
as_msg_make_error_response_bufbuilder(&bmd->keyd, AS_PROTO_RESULT_FAIL_NOTFOUND, bb_r, ns->name);
}
bmd->done = true;
as_partition_release(&rsv);
cf_atomic_int_decr(&g_config.batch_tree_count);
}
else {
cf_debug(AS_BATCH, "batch_build_response: partition reserve read failed: rv %d", rv);
as_msg_make_error_response_bufbuilder(&bmd->keyd, AS_PROTO_RESULT_FAIL_NOTFOUND, bb_r, ns->name);
if (other_node != 0) {
bmd->node = other_node;
cf_debug(AS_BATCH, "other_node is: %p.", other_node);
} else {
cf_debug(AS_BATCH, "other_node is NULL.");
}
}
yield_count++;
if (yield_count % g_config.batch_priority == 0) {
usleep(1);
}
}
}
}
