void
as_record_drop_stats(as_record* r, as_namespace* ns)
{
as_namespace_release_set_id(ns, as_index_get_set_id(r));
cf_atomic64_decr(&ns->n_objects);
}
int
delete_replica(as_partition_reservation* rsv, cf_digest* keyd, bool is_subrec,
bool is_nsup_delete, bool is_xdr_op, cf_node master)
{
// Shortcut pointers & flags.
as_namespace* ns = rsv->ns;
as_index_tree* tree = is_subrec ? rsv->sub_tree : rsv->tree;
as_index_ref r_ref;
r_ref.skip_lock = false;
if (as_record_get(tree, keyd, &r_ref, ns) != 0) {
return AS_PROTO_RESULT_FAIL_NOTFOUND;
}
as_record* r = r_ref.r;
if (ns->storage_data_in_memory) {
as_storage_rd rd;
as_storage_record_open(ns, r, &rd, keyd);
delete_adjust_sindex(&rd);
as_storage_record_close(r, &rd);
}
// Save the set-ID and generation for XDR.
uint16_t set_id = as_index_get_set_id(r);
uint16_t generation = r->generation;
as_index_delete(tree, keyd);
as_record_done(&r_ref, ns);
if (xdr_must_ship_delete(ns, is_nsup_delete, is_xdr_op)) {
xdr_write(ns, *keyd, generation, master, true, set_id, NULL);
}
return AS_PROTO_RESULT_OK;
}
int as_msg_make_response_bufbuilder(as_record *r, as_storage_rd *rd,
cf_buf_builder **bb_r, bool nobindata, char *nsname, bool use_sets,
bool include_key, cf_vector *binlist)
{
// Sanity checks. Either rd should be there or nobindata and nsname should be present.
if (!(rd || (nobindata && nsname))) {
cf_detail(AS_PROTO, "Neither storage record nor nobindata is set. Skipping the record.");
return 0;
}
// figure out the size of the entire buffer
int set_name_len = 0;
const char *set_name = NULL;
int ns_len = rd ? strlen(rd->ns->name) : strlen(nsname);
if (use_sets && as_index_get_set_id(r) != INVALID_SET_ID) {
as_namespace *ns = NULL;
if (rd) {
ns = rd->ns;
} else if (nsname) {
ns = as_namespace_get_byname(nsname);
}
if (!ns) {
cf_info(AS_PROTO, "Cannot get namespace, needed to get set information. Skipping record.");
return -1;
}
set_name = as_index_get_set_name(r, ns);
if (set_name) {
set_name_len = strlen(set_name);
}
}
uint8_t* key = NULL;
uint32_t key_size = 0;
if (include_key && as_index_is_flag_set(r, AS_INDEX_FLAG_KEY_STORED)) {
if (! as_storage_record_get_key(rd)) {
cf_info(AS_PROTO, "can't get key - skipping record");
return -1;
}
key = rd->key;
key_size = rd->key_size;
}
uint16_t n_fields = 2;
int msg_sz = sizeof(as_msg);
msg_sz += sizeof(as_msg_field) + sizeof(cf_digest);
msg_sz += sizeof(as_msg_field) + ns_len;
if (set_name) {
n_fields++;
msg_sz += sizeof(as_msg_field) + set_name_len;
}
if (key) {
n_fields++;
msg_sz += sizeof(as_msg_field) + key_size;
}
int list_bins = 0;
int in_use_bins = 0;
if (rd) {
in_use_bins = as_bin_inuse_count(rd);
}
if (nobindata == false) {
if(binlist) {
int binlist_sz = cf_vector_size(binlist);
for(uint16_t i = 0; i < binlist_sz; i++) {
char binname[AS_ID_BIN_SZ];
cf_vector_get(binlist, i, (void*)&binname);
cf_debug(AS_PROTO, " Binname projected inside is |%s| \n", binname);
as_bin *p_bin = as_bin_get (rd, (uint8_t*)binname, strlen(binname));
if (!p_bin)
{
cf_debug(AS_PROTO, "To be projected bin |%s| not found \n", binname);
continue;
}
cf_debug(AS_PROTO, "Adding bin |%s| to projected bins |%s| \n", binname);
list_bins++;
msg_sz += sizeof(as_msg_op);
msg_sz += rd->ns->single_bin ? 0 : strlen(binname);
uint32_t psz;
if (as_bin_is_hidden(p_bin)) {
psz = 0;
} else {
as_particle_tobuf(p_bin, 0, &psz); // get size
}
msg_sz += psz;
}
}
else {
msg_sz += sizeof(as_msg_op) * in_use_bins; // the bin headers
for (uint16_t i = 0; i < in_use_bins; i++) {
as_bin *p_bin = &rd->bins[i];
msg_sz += rd->ns->single_bin ? 0 : strlen(as_bin_get_name_from_id(rd->ns, p_bin->id));
uint32_t psz;
if (as_bin_is_hidden(p_bin)) {
psz = 0;
} else {
as_particle_tobuf(p_bin, 0, &psz); // get size
}
msg_sz += psz;
}
}
}
uint8_t *b;
cf_buf_builder_reserve(bb_r, msg_sz, &b);
// set up the header
uint8_t *buf = b;
as_msg *msgp = (as_msg *) buf;
msgp->header_sz = sizeof(as_msg);
msgp->info1 = (nobindata ? AS_MSG_INFO1_GET_NOBINDATA : 0);
msgp->info2 = 0;
msgp->info3 = 0;
msgp->unused = 0;
msgp->result_code = 0;
msgp->generation = r->generation;
msgp->record_ttl = r->void_time;
msgp->transaction_ttl = 0;
msgp->n_fields = n_fields;
if (rd) {
if (binlist)
msgp->n_ops = list_bins;
else
msgp->n_ops = in_use_bins;
} else {
msgp->n_ops = 0;
}
as_msg_swap_header(msgp);
buf += sizeof(as_msg);
as_msg_field *mf = (as_msg_field *) buf;
mf->field_sz = sizeof(cf_digest) + 1;
mf->type = AS_MSG_FIELD_TYPE_DIGEST_RIPE;
if (rd) {
memcpy(mf->data, &rd->keyd, sizeof(cf_digest));
} else {
memcpy(mf->data, &r->key, sizeof(cf_digest));
}
as_msg_swap_field(mf);
buf += sizeof(as_msg_field) + sizeof(cf_digest);
mf = (as_msg_field *) buf;
mf->field_sz = ns_len + 1;
mf->type = AS_MSG_FIELD_TYPE_NAMESPACE;
if (rd) {
memcpy(mf->data, rd->ns->name, ns_len);
} else {
memcpy(mf->data, nsname, ns_len);
}
as_msg_swap_field(mf);
buf += sizeof(as_msg_field) + ns_len;
if (set_name) {
mf = (as_msg_field *) buf;
mf->field_sz = set_name_len + 1;
mf->type = AS_MSG_FIELD_TYPE_SET;
memcpy(mf->data, set_name, set_name_len);
as_msg_swap_field(mf);
buf += sizeof(as_msg_field) + set_name_len;
}
if (key) {
mf = (as_msg_field *) buf;
mf->field_sz = key_size + 1;
mf->type = AS_MSG_FIELD_TYPE_KEY;
memcpy(mf->data, key, key_size);
as_msg_swap_field(mf);
buf += sizeof(as_msg_field) + key_size;
}
if (nobindata) {
goto Out;
}
if(binlist) {
int binlist_sz = cf_vector_size(binlist);
for(uint16_t i = 0; i < binlist_sz; i++) {
char binname[AS_ID_BIN_SZ];
cf_vector_get(binlist, i, (void*)&binname);
cf_debug(AS_PROTO, " Binname projected inside is |%s| \n", binname);
as_bin *p_bin = as_bin_get (rd, (uint8_t*)binname, strlen(binname));
if (!p_bin) // should it be checked before ???
continue;
as_msg_op *op = (as_msg_op *)buf;
buf += sizeof(as_msg_op);
op->op = AS_MSG_OP_READ;
op->name_sz = as_bin_memcpy_name(rd->ns, op->name, p_bin);
buf += op->name_sz;
// Since there are two variable bits, the size is everything after
// the data bytes - and this is only the head, we're patching up
// the rest in a minute.
op->op_sz = 4 + op->name_sz;
if (as_bin_inuse(p_bin)) {
op->particle_type = as_particle_type_convert(as_bin_get_particle_type(p_bin));
op->version = as_bin_get_version(p_bin, rd->ns->single_bin);
uint32_t psz = msg_sz - (buf - b); // size remaining in buffer, for safety
if (as_bin_is_hidden(p_bin)) {
op->particle_type = AS_PARTICLE_TYPE_NULL;
psz = 0;
} else {
if (0 != as_particle_tobuf(p_bin, buf, &psz)) {
cf_warning(AS_PROTO, "particle to buf: could not copy data!");
}
}
buf += psz;
op->op_sz += psz;
}
else {
cf_debug(AS_PROTO, "Whoops !! bin not in use");
op->particle_type = AS_PARTICLE_TYPE_NULL;
}
as_msg_swap_op(op);
}
}
else {
// over all bins, copy into the buffer
for (uint16_t i = 0; i < in_use_bins; i++) {
as_msg_op *op = (as_msg_op *)buf;
buf += sizeof(as_msg_op);
op->op = AS_MSG_OP_READ;
op->name_sz = as_bin_memcpy_name(rd->ns, op->name, &rd->bins[i]);
buf += op->name_sz;
// Since there are two variable bits, the size is everything after
// the data bytes - and this is only the head, we're patching up
// the rest in a minute.
op->op_sz = 4 + op->name_sz;
if (as_bin_inuse(&rd->bins[i])) {
op->particle_type = as_particle_type_convert(as_bin_get_particle_type(&rd->bins[i]));
op->version = as_bin_get_version(&rd->bins[i], rd->ns->single_bin);
uint32_t psz = msg_sz - (buf - b); // size remaining in buffer, for safety
if (as_bin_is_hidden(&rd->bins[i])) {
op->particle_type = AS_PARTICLE_TYPE_NULL;
psz = 0;
} else {
if (0 != as_particle_tobuf(&rd->bins[i], buf, &psz)) {
cf_warning(AS_PROTO, "particle to buf: could not copy data!");
}
}
buf += psz;
op->op_sz += psz;
}
else {
op->particle_type = AS_PARTICLE_TYPE_NULL;
}
as_msg_swap_op(op);
}
}
Out:
return(0);
}
// NB: this uses the same logic as the bufbuild function
// as_msg_make_response_bufbuilder() but does not build a buffer and simply
// returns sizing information. This is required for query runtime memory
// accounting.
// returns -1 in case of error
// otherwise returns resize value
size_t as_msg_response_msgsize(as_record *r, as_storage_rd *rd, bool nobindata,
char *nsname, bool use_sets, cf_vector *binlist)
{
// Sanity checks. Either rd should be there or nobindata and nsname should be present.
if (!(rd || (nobindata && nsname))) {
cf_detail(AS_PROTO, "Neither storage record nor nobindata is set. Skipping the record.");
return -1;
}
// figure out the size of the entire buffer
int set_name_len = 0;
const char *set_name = NULL;
int ns_len = rd ? strlen(rd->ns->name) : strlen(nsname);
if (use_sets && as_index_get_set_id(r) != INVALID_SET_ID) {
as_namespace *ns = NULL;
if (rd) {
ns = rd->ns;
} else if (nsname) {
ns = as_namespace_get_byname(nsname);
}
if (!ns) {
cf_info(AS_PROTO, "Cannot get namespace, needed to get set information. Skipping record.");
return -1;
}
set_name = as_index_get_set_name(r, ns);
if (set_name) {
set_name_len = strlen(set_name);
}
}
int msg_sz = sizeof(as_msg);
msg_sz += sizeof(as_msg_field) + sizeof(cf_digest);
msg_sz += sizeof(as_msg_field) + ns_len;
if (set_name) {
msg_sz += sizeof(as_msg_field) + set_name_len;
}
int in_use_bins = as_bin_inuse_count(rd);
int list_bins = 0;
if (nobindata == false) {
if(binlist) {
int binlist_sz = cf_vector_size(binlist);
for(uint16_t i = 0; i < binlist_sz; i++) {
char binname[AS_ID_BIN_SZ];
cf_vector_get(binlist, i, (void*)&binname);
cf_debug(AS_PROTO, " Binname projected inside is |%s| \n", binname);
as_bin *p_bin = as_bin_get (rd, (uint8_t*)binname, strlen(binname));
if (!p_bin)
{
cf_debug(AS_PROTO, "To be projected bin |%s| not found \n", binname);
continue;
}
cf_debug(AS_PROTO, "Adding bin |%s| to projected bins |%s| \n", binname);
list_bins++;
msg_sz += sizeof(as_msg_op);
msg_sz += rd->ns->single_bin ? 0 : strlen(binname);
uint32_t psz;
as_particle_tobuf(p_bin, 0, &psz); // get size
msg_sz += psz;
}
}
else {
msg_sz += sizeof(as_msg_op) * in_use_bins; // the bin headers
for (uint16_t i = 0; i < in_use_bins; i++) {
as_bin *p_bin = &rd->bins[i];
msg_sz += rd->ns->single_bin ? 0 : strlen(as_bin_get_name_from_id(rd->ns, p_bin->id));
uint32_t psz;
as_particle_tobuf(p_bin, 0, &psz); // get size
msg_sz += psz;
}
}
}
return msg_sz;
}
/* Internal Function: Does the post processing for the UDF record after the
* UDF execution. Does the following:
* 1. Record is closed
* 2. urecord_op is updated to delete in case there is no bin left in it.
* 3. record->pickled_buf is populated before the record is close in case
* it was write operation
* 4. UDF updates cache is cleared
*
* Returns: Nothing
*
* Parameters: urecord - UDF record to operate on
* urecord_op (out) - Populated with the optype
*/
void
udf_rw_post_processing(udf_record *urecord, udf_optype *urecord_op, uint16_t set_id)
{
as_storage_rd *rd = urecord->rd;
as_transaction *tr = urecord->tr;
as_index_ref *r_ref = urecord->r_ref;
// INIT
urecord->pickled_buf = NULL;
urecord->pickled_sz = 0;
urecord->pickled_void_time = 0;
as_rec_props_clear(&urecord->pickled_rec_props);
bool udf_xdr_ship_op = false;
// TODO: optimize not to allocate buffer if it is single
// node cluster. No remote to send data to
// Check if UDF has updates.
if (urecord->flag & UDF_RECORD_FLAG_HAS_UPDATES) {
// Check if the record is not deleted after an update
if ( urecord->flag & UDF_RECORD_FLAG_OPEN) {
*urecord_op = UDF_OPTYPE_WRITE;
udf_xdr_ship_op = true;
}
else {
// If the record has updates and it is not open,
// and if it pre-existed it's an update followed by a delete.
if ( urecord->flag & UDF_RECORD_FLAG_PREEXISTS) {
*urecord_op = UDF_OPTYPE_DELETE;
udf_xdr_ship_op = true;
}
// If the record did not pre-exist and is updated
// and it is not open, then it is create followed by
// delete essentially no_op.
else {
*urecord_op = UDF_OPTYPE_NONE;
}
}
} else if ((urecord->flag & UDF_RECORD_FLAG_PREEXISTS)
&& !(urecord->flag & UDF_RECORD_FLAG_OPEN)) {
*urecord_op = UDF_OPTYPE_DELETE;
udf_xdr_ship_op = true;
} else {
*urecord_op = UDF_OPTYPE_READ;
}
cf_detail(AS_UDF, "FINISH working with LDT Record %p %p %p %p %d", &urecord,
urecord->tr, urecord->r_ref, urecord->rd,
(urecord->flag & UDF_RECORD_FLAG_STORAGE_OPEN));
// If there exists a record reference but no bin of the record is in use,
// delete the record. remove from the tree. Only LDT_RECORD here not needed
// for LDT_SUBRECORD (only do it if requested by UDF). All the SUBRECORD of
// removed LDT_RECORD will be lazily cleaned up by defrag.
if (!(urecord->flag & UDF_RECORD_FLAG_IS_SUBRECORD)
&& urecord->flag & UDF_RECORD_FLAG_OPEN
&& !as_bin_inuse_has(rd)) {
as_index_delete(tr->rsv.tree, &tr->keyd);
urecord->starting_memory_bytes = 0;
*urecord_op = UDF_OPTYPE_DELETE;
udf_xdr_ship_op = true;
} else if (*urecord_op == UDF_OPTYPE_WRITE) {
cf_detail(AS_UDF, "Committing Changes %"PRIx64" n_bins %d", rd->keyd, as_bin_get_n_bins(r_ref->r, rd));
size_t rec_props_data_size = as_storage_record_rec_props_size(rd);
uint8_t rec_props_data[rec_props_data_size];
if (rec_props_data_size > 0) {
as_storage_record_set_rec_props(rd, rec_props_data);
}
write_local_post_processing(tr, tr->rsv.ns, NULL, &urecord->pickled_buf,
&urecord->pickled_sz, &urecord->pickled_void_time,
&urecord->pickled_rec_props, true/*increment_generation*/,
NULL, r_ref->r, rd, urecord->starting_memory_bytes);
// Now ok to accommodate a new stored key...
if (! as_index_is_flag_set(r_ref->r, AS_INDEX_FLAG_KEY_STORED) && rd->key) {
if (rd->ns->storage_data_in_memory) {
as_record_allocate_key(r_ref->r, rd->key, rd->key_size);
}
as_index_set_flags(r_ref->r, AS_INDEX_FLAG_KEY_STORED);
}
// ... or drop a stored key.
else if (as_index_is_flag_set(r_ref->r, AS_INDEX_FLAG_KEY_STORED) && ! rd->key) {
if (rd->ns->storage_data_in_memory) {
as_record_remove_key(r_ref->r);
}
as_index_clear_flags(r_ref->r, AS_INDEX_FLAG_KEY_STORED);
}
}
// Collect the record information (for XDR) before closing the record
as_generation generation = 0;
if (urecord->flag & UDF_RECORD_FLAG_OPEN) {
generation = r_ref->r->generation;
set_id = as_index_get_set_id(r_ref->r);
}
// Close the record for all the cases
udf_record_close(urecord, false);
// Write to XDR pipe after closing the record, in order to release the record lock as
// early as possible.
if (udf_xdr_ship_op == true) {
if (UDF_OP_IS_WRITE(*urecord_op)) {
cf_detail(AS_UDF, "UDF write shipping for key %" PRIx64, tr->keyd);
xdr_write(tr->rsv.ns, tr->keyd, generation, 0, false, set_id);
} else if (UDF_OP_IS_DELETE(*urecord_op)) {
cf_detail(AS_UDF, "UDF delete shipping for key %" PRIx64, tr->keyd);
xdr_write(tr->rsv.ns, tr->keyd, generation, 0, true, set_id);
}
}
// Replication happens when the main record replicates
if (urecord->particle_data) {
cf_free(urecord->particle_data);
urecord->particle_data = 0;
}
udf_record_cache_free(urecord);
}
// If remote record is better than local record, replace local with remote.
int
as_record_replace_if_better(as_remote_record *rr, bool is_repl_write,
bool skip_sindex, bool do_xdr_write)
{
as_namespace *ns = rr->rsv->ns;
if (! as_storage_has_space(ns)) {
cf_warning(AS_RECORD, "{%s} record replace: drives full", ns->name);
return AS_ERR_OUT_OF_SPACE;
}
CF_ALLOC_SET_NS_ARENA(ns);
as_index_tree *tree = rr->rsv->tree;
as_index_ref r_ref;
int rv = as_record_get_create(tree, rr->keyd, &r_ref, ns);
if (rv < 0) {
return AS_ERR_OUT_OF_SPACE;
}
bool is_create = rv == 1;
as_index *r = r_ref.r;
int result;
conflict_resolution_pol policy = ns->conflict_resolution_policy;
if (is_repl_write) {
bool from_replica;
if ((result = as_partition_check_source(ns, rr->rsv->p, rr->src,
&from_replica)) != AS_OK) {
record_replace_failed(rr, &r_ref, NULL, is_create);
return result;
}
repl_write_init_repl_state(rr, from_replica);
policy = repl_write_conflict_resolution_policy(ns);
}
if (! is_create && record_replace_check(r, ns) < 0) {
record_replace_failed(rr, &r_ref, NULL, is_create);
return AS_ERR_FORBIDDEN;
}
// If local record is better, no-op or fail.
if (! is_create && (result = as_record_resolve_conflict(policy,
r->generation, r->last_update_time, (uint16_t)rr->generation,
rr->last_update_time)) <= 0) {
record_replace_failed(rr, &r_ref, NULL, is_create);
return result == 0 ? AS_ERR_RECORD_EXISTS : AS_ERR_GENERATION;
}
// else - remote winner - apply it.
// If creating record, write set-ID into index.
if (is_create) {
if (rr->set_name && (result = as_index_set_set_w_len(r, ns,
rr->set_name, rr->set_name_len, false)) < 0) {
record_replace_failed(rr, &r_ref, NULL, is_create);
return -result;
}
r->last_update_time = rr->last_update_time;
// Don't write record if it would be truncated.
if (as_truncate_record_is_truncated(r, ns)) {
record_replace_failed(rr, &r_ref, NULL, is_create);
return AS_OK;
}
}
// else - not bothering to check that sets match.
as_storage_rd rd;
if (is_create) {
as_storage_record_create(ns, r, &rd);
}
else {
as_storage_record_open(ns, r, &rd);
}
// TODO - old pickle - remove condition in "six months".
if (rr->is_old_pickle) {
// Prepare to store set name, if there is one.
rd.set_name = rr->set_name;
rd.set_name_len = rr->set_name_len;
}
else {
rd.pickle = rr->pickle;
rd.pickle_sz = rr->pickle_sz;
rd.orig_pickle_sz = as_flat_orig_pickle_size(rr, rd.pickle_sz);
}
// Note - deal with key after reading existing record (if such), in case
// we're dropping the key.
// Split according to configuration to replace local record.
bool is_delete = false;
if (ns->storage_data_in_memory) {
if (ns->single_bin) {
result = record_apply_dim_single_bin(rr, &rd, &is_delete);
}
else {
result = record_apply_dim(rr, &rd, skip_sindex, &is_delete);
}
}
else {
if (ns->single_bin) {
result = record_apply_ssd_single_bin(rr, &rd, &is_delete);
}
else {
result = record_apply_ssd(rr, &rd, skip_sindex, &is_delete);
}
}
if (result != 0) {
record_replace_failed(rr, &r_ref, &rd, is_create);
return result;
}
uint16_t set_id = as_index_get_set_id(r); // save for XDR write
record_replaced(r, rr);
as_storage_record_close(&rd);
as_record_done(&r_ref, ns);
if (do_xdr_write) {
xdr_write_replica(rr, is_delete, set_id);
}
return AS_OK;
}
int as_msg_make_response_bufbuilder(as_record *r, as_storage_rd *rd,
cf_buf_builder **bb_r, bool nobindata, char *nsname, bool include_ldt_data,
bool include_key, bool skip_empty_records, cf_vector *binlist)
{
// Sanity checks. Either rd should be there or nobindata and nsname should be present.
if (!(rd || (nobindata && nsname))) {
cf_detail(AS_PROTO, "Neither storage record nor nobindata is set. Skipping the record.");
return 0;
}
// figure out the size of the entire buffer
int set_name_len = 0;
const char *set_name = NULL;
int ns_len = rd ? strlen(rd->ns->name) : strlen(nsname);
if (as_index_get_set_id(r) != INVALID_SET_ID) {
as_namespace *ns = NULL;
if (rd) {
ns = rd->ns;
} else if (nsname) {
ns = as_namespace_get_byname(nsname);
}
if (!ns) {
cf_info(AS_PROTO, "Cannot get namespace, needed to get set information. Skipping record.");
return -1;
}
set_name = as_index_get_set_name(r, ns);
if (set_name) {
set_name_len = strlen(set_name);
}
}
uint8_t* key = NULL;
uint32_t key_size = 0;
if (include_key && as_index_is_flag_set(r, AS_INDEX_FLAG_KEY_STORED)) {
if (! as_storage_record_get_key(rd)) {
cf_info(AS_PROTO, "can't get key - skipping record");
return -1;
}
key = rd->key;
key_size = rd->key_size;
}
uint16_t n_fields = 2;
int msg_sz = sizeof(as_msg);
msg_sz += sizeof(as_msg_field) + sizeof(cf_digest);
msg_sz += sizeof(as_msg_field) + ns_len;
if (set_name) {
n_fields++;
msg_sz += sizeof(as_msg_field) + set_name_len;
}
if (key) {
n_fields++;
msg_sz += sizeof(as_msg_field) + key_size;
}
int list_bins = 0;
int in_use_bins = rd ? (int)as_bin_inuse_count(rd) : 0;
as_val *ldt_bin_vals[in_use_bins];
if (! nobindata) {
if (binlist) {
int binlist_sz = cf_vector_size(binlist);
for (uint16_t i = 0; i < binlist_sz; i++) {
char binname[AS_ID_BIN_SZ];
cf_vector_get(binlist, i, (void*)&binname);
as_bin *p_bin = as_bin_get(rd, binname);
if (! p_bin) {
continue;
}
msg_sz += sizeof(as_msg_op);
msg_sz += rd->ns->single_bin ? 0 : strlen(binname);
if (as_bin_is_hidden(p_bin)) {
if (include_ldt_data) {
msg_sz += (int)as_ldt_particle_client_value_size(rd, p_bin, &ldt_bin_vals[list_bins]);
}
else {
ldt_bin_vals[list_bins] = NULL;
}
}
else {
msg_sz += (int)as_bin_particle_client_value_size(p_bin);
}
list_bins++;
}
// Don't return an empty record.
if (skip_empty_records && list_bins == 0) {
return 0;
}
}
else {
msg_sz += sizeof(as_msg_op) * in_use_bins;
for (uint16_t i = 0; i < in_use_bins; i++) {
as_bin *p_bin = &rd->bins[i];
msg_sz += rd->ns->single_bin ? 0 : strlen(as_bin_get_name_from_id(rd->ns, p_bin->id));
if (as_bin_is_hidden(p_bin)) {
if (include_ldt_data) {
msg_sz += (int)as_ldt_particle_client_value_size(rd, p_bin, &ldt_bin_vals[i]);
}
else {
ldt_bin_vals[i] = NULL;
}
}
else {
msg_sz += (int)as_bin_particle_client_value_size(p_bin);
}
}
}
}
uint8_t *b;
cf_buf_builder_reserve(bb_r, msg_sz, &b);
// set up the header
uint8_t *buf = b;
as_msg *msgp = (as_msg *) buf;
msgp->header_sz = sizeof(as_msg);
msgp->info1 = (nobindata ? AS_MSG_INFO1_GET_NOBINDATA : 0);
msgp->info2 = 0;
msgp->info3 = 0;
msgp->unused = 0;
msgp->result_code = 0;
msgp->generation = r->generation;
msgp->record_ttl = r->void_time;
msgp->transaction_ttl = 0;
msgp->n_fields = n_fields;
if (rd) {
if (binlist)
msgp->n_ops = list_bins;
else
msgp->n_ops = in_use_bins;
} else {
msgp->n_ops = 0;
}
as_msg_swap_header(msgp);
buf += sizeof(as_msg);
as_msg_field *mf = (as_msg_field *) buf;
mf->field_sz = sizeof(cf_digest) + 1;
mf->type = AS_MSG_FIELD_TYPE_DIGEST_RIPE;
if (rd) {
memcpy(mf->data, &rd->keyd, sizeof(cf_digest));
} else {
memcpy(mf->data, &r->key, sizeof(cf_digest));
}
as_msg_swap_field(mf);
buf += sizeof(as_msg_field) + sizeof(cf_digest);
mf = (as_msg_field *) buf;
mf->field_sz = ns_len + 1;
mf->type = AS_MSG_FIELD_TYPE_NAMESPACE;
if (rd) {
memcpy(mf->data, rd->ns->name, ns_len);
} else {
memcpy(mf->data, nsname, ns_len);
}
as_msg_swap_field(mf);
buf += sizeof(as_msg_field) + ns_len;
if (set_name) {
mf = (as_msg_field *) buf;
mf->field_sz = set_name_len + 1;
mf->type = AS_MSG_FIELD_TYPE_SET;
memcpy(mf->data, set_name, set_name_len);
as_msg_swap_field(mf);
buf += sizeof(as_msg_field) + set_name_len;
}
if (key) {
mf = (as_msg_field *) buf;
mf->field_sz = key_size + 1;
mf->type = AS_MSG_FIELD_TYPE_KEY;
memcpy(mf->data, key, key_size);
as_msg_swap_field(mf);
buf += sizeof(as_msg_field) + key_size;
}
if (nobindata) {
return 0;
}
if (binlist) {
list_bins = 0;
int binlist_sz = cf_vector_size(binlist);
for (uint16_t i = 0; i < binlist_sz; i++) {
char binname[AS_ID_BIN_SZ];
cf_vector_get(binlist, i, (void*)&binname);
as_bin *p_bin = as_bin_get(rd, binname);
if (! p_bin) {
continue;
}
as_msg_op *op = (as_msg_op *)buf;
op->op = AS_MSG_OP_READ;
op->version = 0;
op->name_sz = as_bin_memcpy_name(rd->ns, op->name, p_bin);
op->op_sz = 4 + op->name_sz;
buf += sizeof(as_msg_op) + op->name_sz;
if (as_bin_is_hidden(p_bin)) {
buf += as_ldt_particle_to_client(ldt_bin_vals[list_bins], op);
}
else {
buf += as_bin_particle_to_client(p_bin, op);
}
list_bins++;
as_msg_swap_op(op);
}
}
else {
for (uint16_t i = 0; i < in_use_bins; i++) {
as_msg_op *op = (as_msg_op *)buf;
op->op = AS_MSG_OP_READ;
op->version = 0;
op->name_sz = as_bin_memcpy_name(rd->ns, op->name, &rd->bins[i]);
op->op_sz = 4 + op->name_sz;
buf += sizeof(as_msg_op) + op->name_sz;
if (as_bin_is_hidden(&rd->bins[i])) {
buf += as_ldt_particle_to_client(ldt_bin_vals[i], op);
}
else {
buf += as_bin_particle_to_client(&rd->bins[i], op);
}
as_msg_swap_op(op);
}
}
return 0;
}
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;
}
/* Internal Function: Does the post processing for the UDF record after the
* UDF execution. Does the following:
* 1. Record is closed
* 2. urecord_op is updated to delete in case there is no bin left in it.
* 3. record->pickled_buf is populated before the record is close in case
* it was write operation
* 4. UDF updates cache is cleared
*
* Returns: Nothing
*
* Parameters: urecord - UDF record to operate on
* urecord_op (out) - Populated with the optype
*/
static void
post_processing(udf_record *urecord, udf_optype *urecord_op, uint16_t set_id)
{
as_storage_rd *rd = urecord->rd;
as_transaction *tr = urecord->tr;
as_index_ref *r_ref = urecord->r_ref;
// INIT
urecord->pickled_buf = NULL;
urecord->pickled_sz = 0;
as_rec_props_clear(&urecord->pickled_rec_props);
bool udf_xdr_ship_op = false;
getop(urecord, urecord_op);
if (UDF_OP_IS_DELETE(*urecord_op)
|| UDF_OP_IS_WRITE(*urecord_op)) {
udf_xdr_ship_op = true;
}
cf_detail(AS_UDF, "FINISH working with LDT Record %p %p %p %p %d", &urecord,
urecord->tr, urecord->r_ref, urecord->rd,
(urecord->flag & UDF_RECORD_FLAG_STORAGE_OPEN));
// If there exists a record reference but no bin of the record is in use,
// delete the record. remove from the tree. Only LDT_RECORD here not needed
// for LDT_SUBRECORD (only do it if requested by UDF). All the SUBRECORD of
// removed LDT_RECORD will be lazily cleaned up by defrag.
if (udf_zero_bins_left(urecord)) {
as_transaction *tr = urecord->tr;
as_index_delete(tr->rsv.tree, &tr->keyd);
urecord->starting_memory_bytes = 0;
*urecord_op = UDF_OPTYPE_DELETE;
}
else if (*urecord_op == UDF_OPTYPE_WRITE) {
cf_detail_digest(AS_UDF, &rd->keyd, "Committing Changes n_bins %d", as_bin_get_n_bins(r_ref->r, rd));
size_t rec_props_data_size = as_storage_record_rec_props_size(rd);
uint8_t rec_props_data[rec_props_data_size];
if (rec_props_data_size > 0) {
as_storage_record_set_rec_props(rd, rec_props_data);
}
write_udf_post_processing(tr, rd, &urecord->pickled_buf,
&urecord->pickled_sz, &urecord->pickled_rec_props,
urecord->starting_memory_bytes);
// Now ok to accommodate a new stored key...
if (! as_index_is_flag_set(r_ref->r, AS_INDEX_FLAG_KEY_STORED) && rd->key) {
if (rd->ns->storage_data_in_memory) {
as_record_allocate_key(r_ref->r, rd->key, rd->key_size);
}
as_index_set_flags(r_ref->r, AS_INDEX_FLAG_KEY_STORED);
}
// ... or drop a stored key.
else if (as_index_is_flag_set(r_ref->r, AS_INDEX_FLAG_KEY_STORED) && ! rd->key) {
if (rd->ns->storage_data_in_memory) {
as_record_remove_key(r_ref->r);
}
as_index_clear_flags(r_ref->r, AS_INDEX_FLAG_KEY_STORED);
}
}
// Collect the record information (for XDR) before closing the record
as_generation generation = 0;
if (urecord->flag & UDF_RECORD_FLAG_OPEN) {
generation = r_ref->r->generation;
set_id = as_index_get_set_id(r_ref->r);
}
urecord->op = *urecord_op;
// Close the record for all the cases
udf_record_close(urecord);
// Write to XDR pipe after closing the record, in order to release the record lock as
// early as possible.
if (udf_xdr_ship_op == true) {
if (UDF_OP_IS_WRITE(*urecord_op)) {
cf_detail(AS_UDF, "UDF write shipping for key %" PRIx64, tr->keyd);
xdr_write(tr->rsv.ns, tr->keyd, generation, 0, false, set_id);
} else if (UDF_OP_IS_DELETE(*urecord_op)) {
cf_detail(AS_UDF, "UDF delete shipping for key %" PRIx64, tr->keyd);
xdr_write(tr->rsv.ns, tr->keyd, generation, 0, true, set_id);
}
}
}
