int ringbuffer_memcpy(ringbuffer_t *dst, ringbuffer_t *src,
lcb_size_t nbytes)
{
ringbuffer_t copy = *src;
struct lcb_iovec_st iov[2];
int ii = 0;
lcb_size_t towrite = nbytes;
lcb_size_t toread, nb;
if (nbytes > ringbuffer_get_nbytes(src)) {
/* EINVAL */
return -1;
}
if (!ringbuffer_ensure_capacity(dst, nbytes)) {
/* Failed to allocate space */
return -1;
}
ringbuffer_get_iov(dst, RINGBUFFER_WRITE, iov);
toread = minimum(iov[ii].iov_len, nbytes);
do {
assert(ii < 2);
nb = ringbuffer_read(©, iov[ii].iov_base, toread);
toread -= nb;
towrite -= nb;
++ii;
} while (towrite > 0);
ringbuffer_produced(dst, nbytes);
return 0;
}
void libcouchbase_server_buffer_start_packet(libcouchbase_server_t *c,
const void *command_cookie,
ringbuffer_t *buff,
ringbuffer_t *buff_cookie,
const void *data,
libcouchbase_size_t size)
{
struct libcouchbase_command_data_st ct;
memset(&ct, 0, sizeof(struct libcouchbase_command_data_st));
/* @todo we don't want to call gethrtime for each operation, */
/* so I need to pass it down the chain so that a large */
/* multiget can reuse the same timer... */
ct.start = gethrtime();
ct.cookie = command_cookie;
if (ringbuffer_get_nbytes(buff_cookie) == 0) {
c->next_timeout = ct.start;
libcouchbase_update_timer(c->instance);
}
if (!ringbuffer_ensure_capacity(buff, size) ||
!ringbuffer_ensure_capacity(&c->cmd_log, size) ||
!ringbuffer_ensure_capacity(buff_cookie, sizeof(ct)) ||
ringbuffer_write(buff, data, size) != size ||
ringbuffer_write(&c->cmd_log, data, size) != size ||
ringbuffer_write(buff_cookie, &ct, sizeof(ct)) != sizeof(ct)) {
abort();
}
}
int ringbuffer_ensure_alignment(ringbuffer_t *c)
{
#if defined(__hpux__) || defined(__hpux) || defined(__sparc__) || defined(__sparc)
intptr_t addr = (intptr_t)c->read_head;
if (addr % 8 != 0) {
ringbuffer_t copy;
if (ringbuffer_initialize(©, c->size) == 0 ||
ringbuffer_memcpy(©, c, ringbuffer_get_nbytes(c)) == -1) {
return -1;
}
ringbuffer_destruct(c);
*c = copy;
}
#else
(void)c;
#endif
return 0;
}
void libcouchbase_server_buffer_retry_packet(libcouchbase_server_t *c,
struct libcouchbase_command_data_st *ct,
ringbuffer_t *buff,
ringbuffer_t *buff_cookie,
const void *data,
libcouchbase_size_t size)
{
libcouchbase_size_t ct_size = sizeof(struct libcouchbase_command_data_st);
if (ringbuffer_get_nbytes(buff_cookie) == 0) {
c->next_timeout = ct->start;
libcouchbase_update_timer(c->instance);
}
if (!ringbuffer_ensure_capacity(buff, size) ||
!ringbuffer_ensure_capacity(&c->cmd_log, size) ||
!ringbuffer_ensure_capacity(buff_cookie, ct_size) ||
ringbuffer_write(buff, data, size) != size ||
ringbuffer_write(&c->cmd_log, data, size) != size ||
ringbuffer_write(buff_cookie, ct, ct_size) != ct_size) {
abort();
}
}
/**
* Extended version of observe command. This allows us to service
* various forms of higher level operations which use observe in one way
* or another
*/
lcb_error_t lcb_observe_ex(lcb_t instance,
const void *command_cookie,
lcb_size_t num,
const void *const *items,
lcb_observe_type_t type)
{
lcb_size_t ii;
lcb_size_t maxix;
lcb_uint32_t opaque;
struct lcb_command_data_st ct;
struct observe_requests_st reqs;
memset(&reqs, 0, sizeof(reqs));
if (instance->type != LCB_TYPE_BUCKET) {
return lcb_synchandler_return(instance, LCB_EBADHANDLE);
}
if (instance->config.handle == NULL) {
return lcb_synchandler_return(instance, LCB_CLIENT_ETMPFAIL);
}
if (instance->config.dist_type != VBUCKET_DISTRIBUTION_VBUCKET) {
return lcb_synchandler_return(instance, LCB_NOT_SUPPORTED);
}
opaque = ++instance->seqno;
ct.cookie = command_cookie;
maxix = instance->config.nreplicas;
if (type == LCB_OBSERVE_TYPE_CHECK) {
maxix = 0;
} else {
if (type == LCB_OBSERVE_TYPE_DURABILITY) {
ct.flags = LCB_CMD_F_OBS_DURABILITY | LCB_CMD_F_OBS_BCAST;
} else {
ct.flags = LCB_CMD_F_OBS_BCAST;
}
}
reqs.nrequests = instance->nservers;
reqs.requests = calloc(reqs.nrequests, sizeof(*reqs.requests));
for (ii = 0; ii < num; ii++) {
const void *key, *hashkey;
lcb_size_t nkey, nhashkey;
int vbid, jj;
if (type == LCB_OBSERVE_TYPE_DURABILITY) {
const lcb_durability_entry_t *ent = items[ii];
key = ent->request.v.v0.key;
nkey = ent->request.v.v0.nkey;
hashkey = ent->request.v.v0.hashkey;
nhashkey = ent->request.v.v0.nhashkey;
} else {
const lcb_observe_cmd_t *ocmd = items[ii];
key = ocmd->v.v0.key;
nkey = ocmd->v.v0.nkey;
hashkey = ocmd->v.v0.hashkey;
nhashkey = ocmd->v.v0.nhashkey;
}
if (!nhashkey) {
hashkey = key;
nhashkey = nkey;
}
vbid = vbucket_get_vbucket_by_key(instance->config.handle,
hashkey, nhashkey);
for (jj = -1; jj < (int)maxix; jj++) {
struct observe_st *rr;
int idx = vbucket_get_replica(instance->config.handle,
vbid, jj);
if (idx < 0 || idx > (int)instance->nservers) {
if (jj == -1) {
destroy_requests(&reqs);
return lcb_synchandler_return(instance, LCB_NO_MATCHING_SERVER);
}
continue;
}
lcb_assert(idx < (int)reqs.nrequests);
rr = reqs.requests + idx;
if (!rr->allocated) {
if (!init_request(rr)) {
destroy_requests(&reqs);
return lcb_synchandler_return(instance, LCB_CLIENT_ENOMEM);
}
}
{
lcb_uint16_t vb = htons((lcb_uint16_t)vbid);
lcb_uint16_t len = htons((lcb_uint16_t)nkey);
rr->packet.message.header.request.magic = PROTOCOL_BINARY_REQ;
rr->packet.message.header.request.opcode = CMD_OBSERVE;
rr->packet.message.header.request.datatype = PROTOCOL_BINARY_RAW_BYTES;
rr->packet.message.header.request.opaque = opaque;
ringbuffer_ensure_capacity(&rr->body,
sizeof(vb) + sizeof(len) + nkey);
rr->nbody += ringbuffer_write(&rr->body, &vb, sizeof(vb));
rr->nbody += ringbuffer_write(&rr->body, &len, sizeof(len));
rr->nbody += ringbuffer_write(&rr->body, key, nkey);
}
}
}
for (ii = 0; ii < reqs.nrequests; ii++) {
struct observe_st *rr = reqs.requests + ii;
struct lcb_server_st *server = instance->servers + ii;
char *tmp;
if (!rr->allocated) {
continue;
}
rr->packet.message.header.request.bodylen = ntohl((lcb_uint32_t)rr->nbody);
ct.start = gethrtime();
lcb_server_start_packet_ct(server, &ct, rr->packet.bytes,
sizeof(rr->packet.bytes));
if (ringbuffer_is_continous(&rr->body, RINGBUFFER_READ, rr->nbody)) {
tmp = ringbuffer_get_read_head(&rr->body);
TRACE_OBSERVE_BEGIN(&rr->packet, server->authority, tmp, rr->nbody);
lcb_server_write_packet(server, tmp, rr->nbody);
} else {
tmp = malloc(ringbuffer_get_nbytes(&rr->body));
if (!tmp) {
/* FIXME by this time some of requests might be scheduled */
destroy_requests(&reqs);
return lcb_synchandler_return(instance, LCB_CLIENT_ENOMEM);
} else {
ringbuffer_read(&rr->body, tmp, rr->nbody);
TRACE_OBSERVE_BEGIN(&rr->packet, server->authority, tmp, rr->nbody);
lcb_server_write_packet(server, tmp, rr->nbody);
}
}
lcb_server_end_packet(server);
lcb_server_send_packets(server);
}
destroy_requests(&reqs);
return lcb_synchandler_return(instance, LCB_SUCCESS);
}
void ringbuffer_reset(ringbuffer_t *buffer)
{
ringbuffer_consumed(buffer,
ringbuffer_get_nbytes(buffer));
}
static void purge_single_server(lcb_server_t *server, lcb_error_t error,
hrtime_t min_nonstale,
hrtime_t *tmo_next)
{
protocol_binary_request_header req;
struct lcb_command_data_st ct;
lcb_size_t nr;
char *packet;
lcb_size_t packetsize;
char *keyptr;
ringbuffer_t rest;
ringbuffer_t *stream = &server->cmd_log;
ringbuffer_t *cookies;
ringbuffer_t *mirror = NULL; /* mirror buffer should be purged with main stream */
lcb_connection_t conn = &server->connection;
lcb_size_t send_size = 0;
lcb_size_t stream_size = ringbuffer_get_nbytes(stream);
hrtime_t now = gethrtime();
if (server->connection_ready) {
cookies = &server->output_cookies;
} else {
cookies = &server->pending_cookies;
mirror = &server->pending;
}
if (conn->output) {
/* This will usually be false for v1 */
send_size = ringbuffer_get_nbytes(conn->output);
}
lcb_assert(ringbuffer_initialize(&rest, 1024));
do {
int allocated = 0;
lcb_uint32_t headersize;
lcb_uint16_t nkey;
nr = ringbuffer_peek(cookies, &ct, sizeof(ct));
if (nr != sizeof(ct)) {
break;
}
nr = ringbuffer_peek(stream, req.bytes, sizeof(req));
if (nr != sizeof(req)) {
break;
}
packetsize = (lcb_uint32_t)sizeof(req) + ntohl(req.request.bodylen);
if (stream->nbytes < packetsize) {
break;
}
if (min_nonstale && ct.start >= min_nonstale) {
lcb_log(LOGARGS(server, INFO),
"Still have %d ms remaining for command",
(ct.start - min_nonstale) / 1000000);
if (tmo_next) {
*tmo_next = (ct.start - min_nonstale) + 1;
}
break;
}
lcb_log(LOGARGS(server, INFO),
"Command with cookie=%p timed out from server %s:%s",
ct.cookie,
server->curhost.host,
server->curhost.port);
ringbuffer_consumed(cookies, sizeof(ct));
lcb_assert(nr == sizeof(req));
packet = stream->read_head;
if (server->instance->histogram) {
lcb_record_metrics(server->instance, now - ct.start,
req.request.opcode);
}
if (server->connection_ready &&
stream_size > send_size && (stream_size - packetsize) < send_size) {
/* Copy the rest of the current packet into the
temporary stream */
/* I do believe I have some IOV functions to do that? */
lcb_size_t nbytes = packetsize - (stream_size - send_size);
lcb_assert(ringbuffer_memcpy(&rest,
conn->output,
nbytes) == 0);
ringbuffer_consumed(conn->output, nbytes);
send_size -= nbytes;
}
stream_size -= packetsize;
headersize = (lcb_uint32_t)sizeof(req) + req.request.extlen + htons(req.request.keylen);
if (!ringbuffer_is_continous(stream, RINGBUFFER_READ, headersize)) {
packet = malloc(headersize);
if (packet == NULL) {
lcb_error_handler(server->instance, LCB_CLIENT_ENOMEM, NULL);
abort();
}
nr = ringbuffer_peek(stream, packet, headersize);
if (nr != headersize) {
lcb_error_handler(server->instance, LCB_EINTERNAL, NULL);
free(packet);
abort();
}
allocated = 1;
}
keyptr = packet + sizeof(req) + req.request.extlen;
nkey = ntohs(req.request.keylen);
failout_single_request(server, &req, &ct, error, keyptr, nkey, packet);
if (allocated) {
free(packet);
}
ringbuffer_consumed(stream, packetsize);
if (mirror) {
ringbuffer_consumed(mirror, packetsize);
}
} while (1); /* CONSTCOND */
if (server->connection_ready && conn->output) {
/* Preserve the rest of the stream */
lcb_size_t nbytes = ringbuffer_get_nbytes(stream);
send_size = ringbuffer_get_nbytes(conn->output);
if (send_size >= nbytes) {
ringbuffer_consumed(conn->output, send_size - nbytes);
lcb_assert(ringbuffer_memcpy(&rest, conn->output, nbytes) == 0);
}
ringbuffer_reset(conn->output);
ringbuffer_append(&rest, conn->output);
}
ringbuffer_destruct(&rest);
lcb_maybe_breakout(server->instance);
}
LIBCOUCHBASE_API
lcb_error_t lcb_observe(lcb_t instance,
const void *command_cookie,
lcb_size_t num,
const lcb_observe_cmd_t *const *items)
{
int vbid, idx, jj;
lcb_size_t ii;
lcb_uint32_t opaque;
struct observe_st *requests;
/* we need a vbucket config before we can start getting data.. */
if (instance->vbucket_config == NULL) {
switch (instance->type) {
case LCB_TYPE_CLUSTER:
return lcb_synchandler_return(instance, LCB_EBADHANDLE);
case LCB_TYPE_BUCKET:
default:
return lcb_synchandler_return(instance, LCB_CLIENT_ETMPFAIL);
}
}
if (instance->dist_type != VBUCKET_DISTRIBUTION_VBUCKET) {
return lcb_synchandler_return(instance, LCB_NOT_SUPPORTED);
}
/* the list of pointers to body buffers for each server */
requests = calloc(instance->nservers, sizeof(struct observe_st));
opaque = ++instance->seqno;
for (ii = 0; ii < num; ++ii) {
const void *key = items[ii]->v.v0.key;
lcb_size_t nkey = items[ii]->v.v0.nkey;
const void *hashkey = items[ii]->v.v0.hashkey;
lcb_size_t nhashkey = items[ii]->v.v0.nhashkey;
if (nhashkey == 0) {
hashkey = key;
nhashkey = nkey;
}
vbid = vbucket_get_vbucket_by_key(instance->vbucket_config, hashkey,
nhashkey);
for (jj = -1; jj < instance->nreplicas; ++jj) {
struct observe_st *rr;
/* it will increment jj to get server index, so (-1 + 1) = 0 (master) */
idx = vbucket_get_replica(instance->vbucket_config, vbid, jj);
if ((idx < 0 || idx > (int)instance->nservers)) {
/* the config says that there is no server yet at that position (-1) */
if (jj == -1) {
/* master node must be available */
destroy_requests(requests, instance->nservers);
return lcb_synchandler_return(instance, LCB_NETWORK_ERROR);
} else {
continue;
}
}
rr = requests + idx;
if (!rr->allocated) {
if (!init_request(rr)) {
destroy_requests(requests, instance->nservers);
return lcb_synchandler_return(instance, LCB_CLIENT_ENOMEM);
}
rr->req.message.header.request.magic = PROTOCOL_BINARY_REQ;
rr->req.message.header.request.opcode = CMD_OBSERVE;
rr->req.message.header.request.datatype = PROTOCOL_BINARY_RAW_BYTES;
rr->req.message.header.request.opaque = opaque;
}
{
lcb_uint16_t vb = htons((lcb_uint16_t)vbid);
lcb_uint16_t len = htons((lcb_uint16_t)nkey);
ringbuffer_ensure_capacity(&rr->body, sizeof(vb) + sizeof(len) + nkey);
rr->nbody += ringbuffer_write(&rr->body, &vb, sizeof(vb));
rr->nbody += ringbuffer_write(&rr->body, &len, sizeof(len));
rr->nbody += ringbuffer_write(&rr->body, key, nkey);
}
}
}
for (ii = 0; ii < instance->nservers; ++ii) {
struct observe_st *rr = requests + ii;
lcb_server_t *server = instance->servers + ii;
if (rr->allocated) {
char *tmp;
rr->req.message.header.request.bodylen = ntohl((lcb_uint32_t)rr->nbody);
lcb_server_start_packet(server, command_cookie, rr->req.bytes, sizeof(rr->req.bytes));
if (ringbuffer_is_continous(&rr->body, RINGBUFFER_READ, rr->nbody)) {
tmp = ringbuffer_get_read_head(&rr->body);
TRACE_OBSERVE_BEGIN(&rr->req, server->authority, tmp, rr->nbody);
lcb_server_write_packet(server, tmp, rr->nbody);
} else {
tmp = malloc(ringbuffer_get_nbytes(&rr->body));
if (!tmp) {
/* FIXME by this time some of requests might be scheduled */
destroy_requests(requests, instance->nservers);
return lcb_synchandler_return(instance, LCB_CLIENT_ENOMEM);
} else {
ringbuffer_read(&rr->body, tmp, rr->nbody);
TRACE_OBSERVE_BEGIN(&rr->req, server->authority, tmp, rr->nbody);
lcb_server_write_packet(server, tmp, rr->nbody);
}
}
lcb_server_end_packet(server);
lcb_server_send_packets(server);
}
}
destroy_requests(requests, instance->nservers);
return lcb_synchandler_return(instance, LCB_SUCCESS);
}
