/**
* Process a slave replication registration message.
*
* We store the various bits of information the slave gives us and generate
* a reply message.
*
* @param router The router instance
* @param slave The slave server
* @param queue The BINLOG_DUMP packet
* @return Non-zero if data was sent
*/
static int
blr_slave_register(ROUTER_INSTANCE *router, ROUTER_SLAVE *slave, GWBUF *queue)
{
GWBUF *resp;
uint8_t *ptr;
int len, slen;
ptr = GWBUF_DATA(queue);
len = extract_field(ptr, 24);
ptr += 4; // Skip length and sequence number
if (*ptr++ != COM_REGISTER_SLAVE)
return 0;
slave->serverid = extract_field(ptr, 32);
ptr += 4;
slen = *ptr++;
if (slen != 0)
{
slave->hostname = strndup((char *)ptr, slen);
ptr += slen;
}
else
slave->hostname = NULL;
slen = *ptr++;
if (slen != 0)
{
ptr += slen;
slave->user = strndup((char *)ptr, slen);
}
else
slave->user = NULL;
slen = *ptr++;
if (slen != 0)
{
slave->passwd = strndup((char *)ptr, slen);
ptr += slen;
}
else
slave->passwd = NULL;
slave->port = extract_field(ptr, 16);
ptr += 2;
slave->rank = extract_field(ptr, 32);
/*
* Now construct a response
*/
if ((resp = gwbuf_alloc(11)) == NULL)
return 0;
ptr = GWBUF_DATA(resp);
encode_value(ptr, 7, 24); // Payload length
ptr += 3;
*ptr++ = 1; // Sequence number
encode_value(ptr, 0, 24);
ptr += 3;
encode_value(ptr, slave->serverid, 32);
slave->state = BLRS_REGISTERED;
return slave->dcb->func.write(slave->dcb, resp);
}
unsigned int NetGameNetworkData::encode_value(unsigned char *d, const NetGameEventValue &value)
{
switch (value.get_type())
{
case NetGameEventValue::null:
*d = 1;
return 1;
case NetGameEventValue::uinteger:
*d = 2;
*reinterpret_cast<unsigned int*>(d + 1) = value.to_uinteger();
return 5;
case NetGameEventValue::integer:
*d = 3;
*reinterpret_cast<int*>(d + 1) = value.to_integer();
return 5;
case NetGameEventValue::number:
*d = 4;
*reinterpret_cast<float*>(d + 1) = value.to_number();
return 5;
case NetGameEventValue::boolean:
*d = value.to_boolean() ? 6 : 5;
return 1;
case NetGameEventValue::string:
{
std::string s = value.to_string();
*d = 7;
*reinterpret_cast<unsigned short*>(d + 1) = s.length();
memcpy(d + 3, s.data(), s.length());
return 3 + s.length();
}
case NetGameEventValue::complex:
{
d[0] = 8;
unsigned l = 1;
for (unsigned int i = 0; i < value.get_member_count(); i++)
l += encode_value(d + l, value.get_member(i));
d[l] = 0;
l++;
return l;
}
case NetGameEventValue::ucharacter:
*d = 9;
*reinterpret_cast<unsigned char*>(d + 1) = value.to_ucharacter();
return 2;
case NetGameEventValue::character:
*d = 10;
*reinterpret_cast<char*>(d + 1) = value.to_character();
return 2;
case NetGameEventValue::binary:
{
DataBuffer s = value.to_binary();
*d = 11;
*reinterpret_cast<unsigned short*>(d + 1) = s.get_size();
memcpy(d + 3, s.get_data(), s.get_size());
return 3 + s.get_size();
}
default:
throw Exception("Unknown game event value type");
}
}
DataBuffer NetGameNetworkData::encode_event(const NetGameEvent &e)
{
unsigned int length = 3 + e.get_name().length();
for (unsigned int i = 0; i < e.get_argument_count(); i++)
length += get_encoded_length(e.get_argument(i));
DataBuffer data(length + 2);
*data.get_data<unsigned short>() = length;
unsigned char *d = data.get_data<unsigned char>() + 2;
// Write name (2 + name length)
unsigned int name_length = e.get_name().length();
*reinterpret_cast<unsigned short*>(d) = name_length;
d += 2;
memcpy(d, e.get_name().data(), name_length);
d += name_length;
for (unsigned int i = 0; i < e.get_argument_count(); i++)
d += encode_value(d, e.get_argument(i));
// Write end marker
*d = 0;
return data;
}
/**
* Send a response to a "SELECT UNIX_TIMESTAMP()" request. This differs from the other
* requests since we do not save a copy of the original interaction with the master
* and simply replay it. We want to always send the current time. We have stored a typcial
* response, which gives us the schema information normally returned. This is sent to the
* client and then we add a dynamic part that will insert the current timestamp data.
* Finally we send a preprepaed EOF packet to end the response stream.
*
* @param router The binlog router instance
* @param slave The slave server to which we are sending the response
* @return Non-zero if data was sent
*/
static int
blr_slave_send_timestamp(ROUTER_INSTANCE *router, ROUTER_SLAVE *slave)
{
GWBUF *pkt;
char timestamp[20];
uint8_t *ptr;
int len, ts_len;
sprintf(timestamp, "%ld", time(0));
ts_len = strlen(timestamp);
len = sizeof(timestamp_def) + sizeof(timestamp_eof) + 5 + ts_len;
if ((pkt = gwbuf_alloc(len)) == NULL)
return 0;
ptr = GWBUF_DATA(pkt);
memcpy(ptr, timestamp_def, sizeof(timestamp_def)); // Fixed preamble
ptr += sizeof(timestamp_def);
encode_value(ptr, ts_len + 1, 24); // Add length of data packet
ptr += 3;
*ptr++ = 0x04; // Sequence number in response
*ptr++ = ts_len; // Length of result string
strncpy((char *)ptr, timestamp, ts_len); // Result string
ptr += ts_len;
memcpy(ptr, timestamp_eof, sizeof(timestamp_eof)); // EOF packet to terminate result
return slave->dcb->func.write(slave->dcb, pkt);
}
string t_header::encode_env(void) const {
string s("SIP_");
s += toupper(replace_char(header_name, '-', '_'));
s += '=';
s += encode_value();
return s;
}
/**
* Send a "fake" format description event to the newly connected slave
*
* @param router The router instance
* @param slave The slave to send the event to
*/
static void
blr_slave_send_fde(ROUTER_INSTANCE *router, ROUTER_SLAVE *slave)
{
BLFILE *file;
REP_HEADER hdr;
GWBUF *record, *head;
uint8_t *ptr;
uint32_t chksum;
if ((file = blr_open_binlog(router, slave->binlogfile)) == NULL)
return;
if ((record = blr_read_binlog(router, file, 4, &hdr)) == NULL)
{
blr_close_binlog(router, file);
return;
}
blr_close_binlog(router, file);
head = gwbuf_alloc(5);
ptr = GWBUF_DATA(head);
encode_value(ptr, hdr.event_size + 1, 24); // Payload length
ptr += 3;
*ptr++ = slave->seqno++;
*ptr++ = 0; // OK
head = gwbuf_append(head, record);
ptr = GWBUF_DATA(record);
encode_value(ptr, time(0), 32); // Overwrite timestamp
ptr += 13;
encode_value(ptr, 0, 32); // Set next position to 0
/*
* Since we have changed the timestamp we must recalculate the CRC
*
* Position ptr to the start of the event header,
* calculate a new checksum
* and write it into the header
*/
ptr = GWBUF_DATA(record) + hdr.event_size - 4;
chksum = crc32(0L, NULL, 0);
chksum = crc32(chksum, GWBUF_DATA(record), hdr.event_size - 4);
encode_value(ptr, chksum, 32);
slave->dcb->func.write(slave->dcb, head);
}
/**
* Populate a header structure for a replication message from a GWBUF structure.
*
* @param pkt The incoming packet in a GWBUF chain
* @param hdr The packet header to populate
* @return A pointer to the first byte following the event header
*/
uint8_t *
blr_build_header(GWBUF *pkt, REP_HEADER *hdr)
{
uint8_t *ptr;
ptr = GWBUF_DATA(pkt);
encode_value(ptr, hdr->payload_len, 24);
ptr += 3;
*ptr++ = hdr->seqno;
*ptr++ = hdr->ok;
encode_value(ptr, hdr->timestamp, 32);
ptr += 4;
*ptr++ = hdr->event_type;
encode_value(ptr, hdr->serverid, 32);
ptr += 4;
encode_value(ptr, hdr->event_size, 32);
ptr += 4;
encode_value(ptr, hdr->next_pos, 32);
ptr += 4;
encode_value(ptr, hdr->flags, 16);
ptr += 2;
return ptr;
}
/**
* Send the column definition packet in a response packet sequence.
*
* @param router The router
* @param slave The slave connection
* @param name Name of the column
* @param type Column type
* @param len Column length
* @param seqno Packet sequence number
* @return Non-zero on success
*/
static int
blr_slave_send_columndef(ROUTER_INSTANCE *router, ROUTER_SLAVE *slave, char *name, int type, int len, uint8_t seqno)
{
GWBUF *pkt;
uint8_t *ptr;
if ((pkt = gwbuf_alloc(26 + strlen(name))) == NULL)
return 0;
ptr = GWBUF_DATA(pkt);
encode_value(ptr, 22 + strlen(name), 24); // Add length of data packet
ptr += 3;
*ptr++ = seqno; // Sequence number in response
*ptr++ = 3; // Catalog is always def
*ptr++ = 'd';
*ptr++ = 'e';
*ptr++ = 'f';
*ptr++ = 0; // Schema name length
*ptr++ = 0; // virtal table name length
*ptr++ = 0; // Table name length
*ptr++ = strlen(name); // Column name length;
while (*name)
*ptr++ = *name++; // Copy the column name
*ptr++ = 0; // Orginal column name
*ptr++ = 0x0c; // Length of next fields always 12
*ptr++ = 0x3f; // Character set
*ptr++ = 0;
encode_value(ptr, len, 32); // Add length of column
ptr += 4;
*ptr++ = type;
*ptr++ = 0x81; // Two bytes of flags
if (type == 0xfd)
*ptr++ = 0x1f;
else
*ptr++ = 0x00;
*ptr++= 0;
*ptr++= 0;
*ptr++= 0;
return slave->dcb->func.write(slave->dcb, pkt);
}
string t_header::encode(void) const {
string s;
if (!populated) return s;
s = (t_parser::compact_headers && !compact_name.empty() ?
compact_name : header_name);
s += ": ";
s += encode_value();
s += CRLF;
return s;
}
/**
* Send the field count packet in a response packet sequence.
*
* @param router The router
* @param slave The slave connection
* @param count Number of columns in the result set
* @return Non-zero on success
*/
static int
blr_slave_send_fieldcount(ROUTER_INSTANCE *router, ROUTER_SLAVE *slave, int count)
{
GWBUF *pkt;
uint8_t *ptr;
if ((pkt = gwbuf_alloc(5)) == NULL)
return 0;
ptr = GWBUF_DATA(pkt);
encode_value(ptr, 1, 24); // Add length of data packet
ptr += 3;
*ptr++ = 0x01; // Sequence number in response
*ptr++ = count; // Length of result string
return slave->dcb->func.write(slave->dcb, pkt);
}
static void
cb_params_store_init_item(struct params_st *params, lcb_size_t idx,
VALUE key_obj, VALUE value_obj, lcb_uint32_t flags, lcb_cas_t cas,
lcb_time_t exptime)
{
key_obj = unify_key(params->bucket, key_obj, 1);
value_obj = encode_value(value_obj, params->cmd.store.flags);
if (value_obj == Qundef) {
rb_raise(eValueFormatError, "unable to convert value for key '%s'", RSTRING_PTR(key_obj));
}
params->cmd.store.items[idx].v.v0.datatype = params->cmd.store.datatype;
params->cmd.store.items[idx].v.v0.operation = params->cmd.store.operation;
params->cmd.store.items[idx].v.v0.key = RSTRING_PTR(key_obj);
params->cmd.store.items[idx].v.v0.nkey = RSTRING_LEN(key_obj);
params->cmd.store.items[idx].v.v0.bytes = RSTRING_PTR(value_obj);
params->cmd.store.items[idx].v.v0.nbytes = RSTRING_LEN(value_obj);
params->cmd.store.items[idx].v.v0.flags = flags;
params->cmd.store.items[idx].v.v0.cas = cas;
params->cmd.store.items[idx].v.v0.exptime = exptime;
params->npayload += RSTRING_LEN(key_obj) + RSTRING_LEN(value_obj) + sizeof(flags) + sizeof(exptime);
}
/**
* Construct an error response
*
* @param router The router instance
* @param slave The slave server instance
* @param msg The error message to send
*/
static void
blr_slave_send_error(ROUTER_INSTANCE *router, ROUTER_SLAVE *slave, char *msg)
{
GWBUF *pkt;
unsigned char *data;
int len;
if ((pkt = gwbuf_alloc(strlen(msg) + 13)) == NULL)
return;
data = GWBUF_DATA(pkt);
len = strlen(msg) + 1;
encode_value(&data[0], len, 24); // Payload length
data[3] = 0; // Sequence id
// Payload
data[4] = 0xff; // Error indicator
data[5] = 0; // Error Code
data[6] = 0; // Error Code
strncpy((char *)&data[7], "#00000", 6);
memcpy(&data[13], msg, strlen(msg)); // Error Message
slave->dcb->func.write(slave->dcb, pkt);
}
// Return the data for all the names[], start .. end etc.
// as get_values() is supposed to return them.
//
// Returns raw values if interpol <= 0.0.
// Returns 0 on error.
xmlrpc_value *get_sheet_data(xmlrpc_env *env,
int key,
const stdVector<stdString> names,
const epicsTime &start, const epicsTime &end,
long count,
ReaderFactory::How how, double delta)
{
try
{
#ifdef LOGFILE
stdString txt;
LOG_MSG("get_sheet_data\n");
LOG_MSG("Start : %s\n", epicsTimeTxt(start, txt));
LOG_MSG("End : %s\n", epicsTimeTxt(end, txt));
LOG_MSG("Method: %s\n", ReaderFactory::toString(how, delta));
#endif
AutoPtr<Index> index(open_index(env, key));
if (env->fault_occurred)
return 0;
AutoPtr<SpreadsheetReader> sheet(new SpreadsheetReader(*index, how, delta));
AutoXmlRpcValue results(xmlrpc_build_value(env, "()"));
if (env->fault_occurred)
return 0;
long name_count = names.size();
AutoArrayPtr<AutoXmlRpcValue> meta (new AutoXmlRpcValue[name_count]);
AutoArrayPtr<AutoXmlRpcValue> values (new AutoXmlRpcValue[name_count]);
AutoArrayPtr<xmlrpc_int32> xml_type (new xmlrpc_int32[name_count]);
AutoArrayPtr<xmlrpc_int32> xml_count(new xmlrpc_int32[name_count]);
AutoArrayPtr<size_t> ch_vals (new size_t[name_count]);
bool ok = sheet->find(names, &start);
long i;
// Per-channel meta-info.
for (i=0; i<name_count; ++i)
{
#ifdef LOGFILE
LOG_MSG("Handling '%s'\n", names[i].c_str());
#endif
ch_vals[i] = 0;
values[i] = xmlrpc_build_value(env, "()");
if (env->fault_occurred)
return 0;
if (sheet->found(i))
{ // Fix meta/type/count based on first value
meta[i] = encode_ctrl_info(env, &sheet->getInfo(i));
dbr_type_to_xml_type(sheet->getType(i), sheet->getCount(i),
xml_type[i], xml_count[i]);
#if 0
LOG_MSG("Ch %lu: type, count = %d, %d\n", i, (int)xml_type[i], (int)xml_count[i]);
#endif
}
else
{ // Channel exists, but has no data
meta[i] = encode_ctrl_info(env, 0);
xml_type[i] = XML_ENUM;
xml_count[i] = 1;
}
}
// Collect values
long num_vals = 0;
while (ok && num_vals < count && sheet->getTime() < end)
{
for (i=0; i<name_count; ++i)
{
if (sheet->get(i))
{
++ch_vals[i];
encode_value(env,
sheet->getType(i), sheet->getCount(i),
sheet->getTime(), sheet->get(i),
xml_type[i], xml_count[i], values[i]);
}
else
{ // Encode as no value, but one of them ;-)
// to avoid confusion w/ viewers that like to see some data in any case.
encode_value(env, 0, 1, sheet->getTime(), 0,
xml_type[i], xml_count[i], values[i]);
}
}
++num_vals;
ok = sheet->next();
}
// Assemble result = { name, meta, type, count, values }
for (i=0; i<name_count; ++i)
{
AutoXmlRpcValue result(
xmlrpc_build_value(env, "{s:s,s:V,s:i,s:i,s:V}",
"name", names[i].c_str(),
"meta", meta[i].get(),
"type", xml_type[i],
"count", xml_count[i],
"values", values[i].get()));
// Add to result array
xmlrpc_array_append_item(env, results, result);
#ifdef LOGFILE
LOG_MSG("Ch %lu: %zu values\n", i, ch_vals[i]);
#endif
}
#ifdef LOGFILE
LOG_MSG("%ld values total\n", num_vals);
#endif
return results.release();
}
catch (GenericException &e)
{
#ifdef LOGFILE
LOG_MSG("Error:\n%s\n", e.what());
#endif
xmlrpc_env_set_fault_formatted(env, ARCH_DAT_DATA_ERROR,
"%s", e.what());
}
return 0;
}
// Return the data for all the names[], start .. end etc.
// as get_values() is supposed to return them.
// Returns 0 on error.
xmlrpc_value *get_channel_data(xmlrpc_env *env,
int key,
const stdVector<stdString> names,
const epicsTime &start, const epicsTime &end,
long count,
ReaderFactory::How how, double delta)
{
AutoXmlRpcValue results;
try
{
#ifdef LOGFILE
stdString txt;
LOG_MSG("get_channel_data\n");
LOG_MSG("Method: %s\n", ReaderFactory::toString(how, delta));
LOG_MSG("Start: %s\n", epicsTimeTxt(start, txt));
LOG_MSG("End : %s\n", epicsTimeTxt(end, txt));
#endif
AutoPtr<Index> index(open_index(env, key));
if (env->fault_occurred)
return 0;
AutoPtr<DataReader> reader(ReaderFactory::create(*index, how, delta));
results = xmlrpc_build_value(env, "()");
if (env->fault_occurred)
return 0;
long i, name_count = names.size();
for (i=0; i<name_count; ++i)
{
#ifdef LOGFILE
LOG_MSG("Handling '%s'\n", names[i].c_str());
#endif
long num_vals = 0;
AutoXmlRpcValue values(xmlrpc_build_value(env, "()"));
if (env->fault_occurred)
return 0;
const RawValue::Data *data = reader->find(names[i], &start);
AutoXmlRpcValue meta;
xmlrpc_int32 xml_type, xml_count;
if (data == 0)
{ // No exception from file error etc., just no data.
meta = encode_ctrl_info(env, 0);
xml_type = XML_ENUM;
xml_count = 1;
}
else
{
// Fix meta/type/count based on first value
meta = encode_ctrl_info(env, &reader->getInfo());
dbr_type_to_xml_type(reader->getType(), reader->getCount(),
xml_type, xml_count);
while (num_vals < count
&& data
&& RawValue::getTime(data) < end)
{
encode_value(env, reader->getType(), reader->getCount(),
RawValue::getTime(data), data,
xml_type, xml_count, values);
++num_vals;
data = reader->next();
}
}
// Assemble result = { name, meta, type, count, values }
AutoXmlRpcValue result(xmlrpc_build_value(
env, "{s:s,s:V,s:i,s:i,s:V}",
"name", names[i].c_str(),
"meta", (xmlrpc_value *)meta,
"type", xml_type,
"count", xml_count,
"values", (xmlrpc_value *)values));
// Add to result array
xmlrpc_array_append_item(env, results, result);
#ifdef LOGFILE
LOG_MSG("%ld values\n", num_vals);
#endif
} // for ( .. name .. )
}
catch (GenericException &e)
{
#ifdef LOGFILE
LOG_MSG("Error:\n%s\n", e.what());
#endif
xmlrpc_env_set_fault_formatted(env, ARCH_DAT_DATA_ERROR,
"%s", e.what());
return 0;
}
return results.release();
}
static int load_registration_id(PROV_AUTH_INFO* handle)
{
int result;
if (handle->sec_type == PROV_AUTH_TYPE_TPM)
{
SHA256Context sha_ctx;
uint8_t msg_digest[SHA256HashSize];
unsigned char* endorsement_key;
size_t ek_len;
if (handle->hsm_client_get_endorsement_key(handle->hsm_client_handle, &endorsement_key, &ek_len) != 0)
{
LogError("Failed getting device reg id");
result = MU_FAILURE;
}
else
{
if (SHA256Reset(&sha_ctx) != 0)
{
LogError("Failed sha256 reset");
result = MU_FAILURE;
}
else if (SHA256Input(&sha_ctx, endorsement_key, (unsigned int)ek_len) != 0)
{
LogError("Failed SHA256Input");
result = MU_FAILURE;
}
else if (SHA256Result(&sha_ctx, msg_digest) != 0)
{
LogError("Failed SHA256Result");
result = MU_FAILURE;
}
else
{
handle->registration_id = encode_value(msg_digest, SHA256HashSize);
if (handle->registration_id == NULL)
{
LogError("Failed allocating registration Id");
result = MU_FAILURE;
}
else
{
result = 0;
}
}
free(endorsement_key);
}
}
else
{
handle->registration_id = handle->hsm_client_get_common_name(handle->hsm_client_handle);
if (handle->registration_id == NULL)
{
LogError("Failed getting common name from certificate");
result = MU_FAILURE;
}
else
{
result = 0;
}
}
return result;
}
/**
* Generate an internal rotate event that we can use to cause the slave to move beyond
* a binlog file that is misisng the rotate eent at the end.
*
* @param router The router instance
* @param slave The slave to rotate
* @return Non-zero if the rotate took place
*/
static int
blr_slave_fake_rotate(ROUTER_INSTANCE *router, ROUTER_SLAVE *slave)
{
char *sptr;
int filenum;
GWBUF *resp;
uint8_t *ptr;
int len, binlognamelen;
REP_HEADER hdr;
uint32_t chksum;
if ((sptr = strrchr(slave->binlogfile, '.')) == NULL)
return 0;
blr_close_binlog(router, slave->file);
filenum = atoi(sptr + 1);
sprintf(slave->binlogfile, BINLOG_NAMEFMT, router->fileroot, filenum + 1);
slave->binlog_pos = 4;
if ((slave->file = blr_open_binlog(router, slave->binlogfile)) == NULL)
return 0;
binlognamelen = strlen(slave->binlogfile);
if (slave->nocrc)
len = 19 + 8 + binlognamelen;
else
len = 19 + 8 + 4 + binlognamelen;
// Build a fake rotate event
resp = gwbuf_alloc(len + 5);
hdr.payload_len = len + 1;
hdr.seqno = slave->seqno++;
hdr.ok = 0;
hdr.timestamp = 0L;
hdr.event_type = ROTATE_EVENT;
hdr.serverid = router->masterid;
hdr.event_size = len;
hdr.next_pos = 0;
hdr.flags = 0x20;
ptr = blr_build_header(resp, &hdr);
encode_value(ptr, slave->binlog_pos, 64);
ptr += 8;
memcpy(ptr, slave->binlogfile, binlognamelen);
ptr += binlognamelen;
if (!slave->nocrc)
{
/*
* Now add the CRC to the fake binlog rotate event.
*
* The algorithm is first to compute the checksum of an empty buffer
* and then the checksum of the event portion of the message, ie we do not
* include the length, sequence number and ok byte that makes up the first
* 5 bytes of the message. We also do not include the 4 byte checksum itself.
*/
chksum = crc32(0L, NULL, 0);
chksum = crc32(chksum, GWBUF_DATA(resp) + 5, hdr.event_size - 4);
encode_value(ptr, chksum, 32);
}
slave->dcb->func.write(slave->dcb, resp);
return 1;
}
/**
* We have a registered slave that is behind the current leading edge of the
* binlog. We must replay the log entries to bring this node up to speed.
*
* There may be a large number of records to send to the slave, the process
* is triggered by the slave COM_BINLOG_DUMP message and all the events must
* be sent without receiving any new event. This measn there is no trigger into
* MaxScale other than this initial message. However, if we simply send all the
* events we end up with an extremely long write queue on the DCB and risk
* running the server out of resources.
*
* The slave catchup routine will send a burst of replication events per single
* call. The paramter "long" control the number of events in the burst. The
* short burst is intended to be used when the master receive an event and
* needs to put the slave into catchup mode. This prevents the slave taking
* too much tiem away from the thread that is processing the master events.
*
* At the end of the burst a fake EPOLLOUT event is added to the poll event
* queue. This ensures that the slave callback for processing DCB write drain
* will be called and future catchup requests will be handled on another thread.
*
* @param router The binlog router
* @param slave The slave that is behind
* @param large Send a long or short burst of events
* @return The number of bytes written
*/
int
blr_slave_catchup(ROUTER_INSTANCE *router, ROUTER_SLAVE *slave, bool large)
{
GWBUF *head, *record;
REP_HEADER hdr;
int written, rval = 1, burst;
int rotating;
unsigned long burst_size;
uint8_t *ptr;
if (large)
burst = router->long_burst;
else
burst = router->short_burst;
burst_size = router->burst_size;
spinlock_acquire(&slave->catch_lock);
if (slave->cstate & CS_BUSY)
{
spinlock_release(&slave->catch_lock);
return 0;
}
slave->cstate |= CS_BUSY;
spinlock_release(&slave->catch_lock);
if (slave->file == NULL)
{
rotating = router->rotating;
if ((slave->file = blr_open_binlog(router, slave->binlogfile)) == NULL)
{
if (rotating)
{
spinlock_acquire(&slave->catch_lock);
slave->cstate |= CS_EXPECTCB;
slave->cstate &= ~CS_BUSY;
spinlock_release(&slave->catch_lock);
poll_fake_write_event(slave->dcb);
return rval;
}
LOGIF(LE, (skygw_log_write(
LOGFILE_ERROR,
"blr_slave_catchup failed to open binlog file %s",
slave->binlogfile)));
slave->cstate &= ~CS_BUSY;
slave->state = BLRS_ERRORED;
dcb_close(slave->dcb);
return 0;
}
}
slave->stats.n_bursts++;
while (burst-- && burst_size > 0 &&
(record = blr_read_binlog(router, slave->file, slave->binlog_pos, &hdr)) != NULL)
{
head = gwbuf_alloc(5);
ptr = GWBUF_DATA(head);
encode_value(ptr, hdr.event_size + 1, 24);
ptr += 3;
*ptr++ = slave->seqno++;
*ptr++ = 0; // OK
head = gwbuf_append(head, record);
if (hdr.event_type == ROTATE_EVENT)
{
unsigned long beat1 = hkheartbeat;
blr_close_binlog(router, slave->file);
if (hkheartbeat - beat1 > 1) LOGIF(LE, (skygw_log_write(
LOGFILE_ERROR, "blr_close_binlog took %d beats",
hkheartbeat - beat1)));
blr_slave_rotate(slave, GWBUF_DATA(record));
beat1 = hkheartbeat;
if ((slave->file = blr_open_binlog(router, slave->binlogfile)) == NULL)
{
if (rotating)
{
spinlock_acquire(&slave->catch_lock);
slave->cstate |= CS_EXPECTCB;
slave->cstate &= ~CS_BUSY;
spinlock_release(&slave->catch_lock);
poll_fake_write_event(slave->dcb);
return rval;
}
LOGIF(LE, (skygw_log_write(
LOGFILE_ERROR,
"blr_slave_catchup failed to open binlog file %s",
slave->binlogfile)));
slave->state = BLRS_ERRORED;
dcb_close(slave->dcb);
break;
}
if (hkheartbeat - beat1 > 1) LOGIF(LE, (skygw_log_write(
LOGFILE_ERROR, "blr_open_binlog took %d beats",
hkheartbeat - beat1)));
}
slave->stats.n_bytes += gwbuf_length(head);
written = slave->dcb->func.write(slave->dcb, head);
if (written && hdr.event_type != ROTATE_EVENT)
{
slave->binlog_pos = hdr.next_pos;
}
rval = written;
slave->stats.n_events++;
burst_size -= hdr.event_size;
}
if (record == NULL)
slave->stats.n_failed_read++;
spinlock_acquire(&slave->catch_lock);
slave->cstate &= ~CS_BUSY;
spinlock_release(&slave->catch_lock);
if (record)
{
slave->stats.n_flows++;
spinlock_acquire(&slave->catch_lock);
slave->cstate |= CS_EXPECTCB;
spinlock_release(&slave->catch_lock);
poll_fake_write_event(slave->dcb);
}
else if (slave->binlog_pos == router->binlog_position &&
strcmp(slave->binlogfile, router->binlog_name) == 0)
{
int state_change = 0;
spinlock_acquire(&router->binlog_lock);
spinlock_acquire(&slave->catch_lock);
/*
* Now check again since we hold the router->binlog_lock
* and slave->catch_lock.
*/
if (slave->binlog_pos != router->binlog_position ||
strcmp(slave->binlogfile, router->binlog_name) != 0)
{
slave->cstate &= ~CS_UPTODATE;
slave->cstate |= CS_EXPECTCB;
spinlock_release(&slave->catch_lock);
spinlock_release(&router->binlog_lock);
poll_fake_write_event(slave->dcb);
}
else
{
if ((slave->cstate & CS_UPTODATE) == 0)
{
slave->stats.n_upd++;
slave->cstate |= CS_UPTODATE;
spinlock_release(&slave->catch_lock);
spinlock_release(&router->binlog_lock);
state_change = 1;
}
}
if (state_change)
{
slave->stats.n_caughtup++;
if (slave->stats.n_caughtup == 1)
{
LOGIF(LM, (skygw_log_write(LOGFILE_MESSAGE,
"%s: Slave %s is up to date %s, %u.",
router->service->name,
slave->dcb->remote,
slave->binlogfile, slave->binlog_pos)));
}
else if ((slave->stats.n_caughtup % 50) == 0)
{
LOGIF(LM, (skygw_log_write(LOGFILE_MESSAGE,
"%s: Slave %s is up to date %s, %u.",
router->service->name,
slave->dcb->remote,
slave->binlogfile, slave->binlog_pos)));
}
}
}
else
{
if (slave->binlog_pos >= blr_file_size(slave->file)
&& router->rotating == 0
&& strcmp(router->binlog_name, slave->binlogfile) != 0
&& blr_master_connected(router))
{
/* We may have reached the end of file of a non-current
* binlog file.
*
* Note if the master is rotating there is a window during
* which the rotate event has been written to the old binlog
* but the new binlog file has not yet been created. Therefore
* we ignore these issues during the rotate processing.
*/
LOGIF(LE, (skygw_log_write(LOGFILE_ERROR,
"Slave reached end of file for binlong file %s at %u "
"which is not the file currently being downloaded. "
"Master binlog is %s, %lu.",
slave->binlogfile, slave->binlog_pos,
router->binlog_name, router->binlog_position)));
if (blr_slave_fake_rotate(router, slave))
{
spinlock_acquire(&slave->catch_lock);
slave->cstate |= CS_EXPECTCB;
spinlock_release(&slave->catch_lock);
poll_fake_write_event(slave->dcb);
}
else
{
slave->state = BLRS_ERRORED;
dcb_close(slave->dcb);
}
}
else
{
spinlock_acquire(&slave->catch_lock);
slave->cstate |= CS_EXPECTCB;
spinlock_release(&slave->catch_lock);
poll_fake_write_event(slave->dcb);
}
}
return rval;
}
void encode_key(Buffer& buf, const KeyObject& key)
{
uint32 header = (uint32) (key.db << 8) + key.type;
BufferHelper::WriteFixUInt32(buf, header);
BufferHelper::WriteVarSlice(buf, key.key);
switch (key.type)
{
case HASH_FIELD:
{
const HashKeyObject& hk = (const HashKeyObject&) key;
BufferHelper::WriteVarSlice(buf, hk.field);
break;
}
case LIST_ELEMENT:
{
const ListKeyObject& lk = (const ListKeyObject&) key;
BufferHelper::WriteFixFloat(buf, lk.score);
break;
}
case SET_ELEMENT:
{
const SetKeyObject& sk = (const SetKeyObject&) key;
encode_value(buf, sk.value);
break;
}
case ZSET_ELEMENT:
{
const ZSetKeyObject& sk = (const ZSetKeyObject&) key;
BufferHelper::WriteFixDouble(buf, sk.score);
encode_value(buf, sk.value);
break;
}
case ZSET_ELEMENT_SCORE:
{
const ZSetScoreKeyObject& zk = (const ZSetScoreKeyObject&) key;
encode_value(buf, zk.value);
break;
}
case TABLE_INDEX:
{
const TableIndexKeyObject& index =
(const TableIndexKeyObject&) key;
BufferHelper::WriteVarSlice(buf, index.colname);
encode_value(buf, index.colvalue);
BufferHelper::WriteVarUInt32(buf, index.index.size());
ValueArray::const_iterator it = index.index.begin();
while (it != index.index.end())
{
encode_value(buf, *it);
it++;
}
break;
}
case TABLE_COL:
{
const TableColKeyObject& col = (const TableColKeyObject&) key;
BufferHelper::WriteVarSlice(buf, col.colname);
BufferHelper::WriteVarUInt32(buf, col.index.size());
ValueArray::const_iterator it = col.index.begin();
while (it != col.index.end())
{
encode_value(buf, *it);
it++;
}
break;
}
case BITSET_ELEMENT:
{
const BitSetKeyObject& bk = (const BitSetKeyObject&) key;
BufferHelper::WriteVarUInt64(buf, bk.index);
break;
}
case KEY_EXPIRATION_ELEMENT:
{
const ExpireKeyObject& bk = (const ExpireKeyObject&) key;
BufferHelper::WriteVarUInt64(buf, bk.expireat);
break;
}
case LIST_META:
case ZSET_META:
case SET_META:
case TABLE_META:
case TABLE_SCHEMA:
case BITSET_META:
case KEY_EXPIRATION_MAPPING:
case SCRIPT:
default:
{
break;
}
}
}
/**
* Process a COM_BINLOG_DUMP message from the slave. This is the
* final step in the process of registration. The new master, MaxScale
* must send a response packet and generate a fake BINLOG_ROTATE event
* with the binlog file requested by the slave. And then send a
* FORMAT_DESCRIPTION_EVENT that has been saved from the real master.
*
* Once send MaxScale must continue to send binlog events to the slave.
*
* @param router The router instance
* @param slave The slave server
* @param queue The BINLOG_DUMP packet
* @return The number of bytes written to the slave
*/
static int
blr_slave_binlog_dump(ROUTER_INSTANCE *router, ROUTER_SLAVE *slave, GWBUF *queue)
{
GWBUF *resp;
uint8_t *ptr;
int len, flags, serverid, rval, binlognamelen;
REP_HEADER hdr;
uint32_t chksum;
ptr = GWBUF_DATA(queue);
len = extract_field(ptr, 24);
binlognamelen = len - 11;
ptr += 4; // Skip length and sequence number
if (*ptr++ != COM_BINLOG_DUMP)
{
LOGIF(LE, (skygw_log_write(
LOGFILE_ERROR,
"blr_slave_binlog_dump expected a COM_BINLOG_DUMP but received %d",
*(ptr-1))));
return 0;
}
slave->binlog_pos = extract_field(ptr, 32);
ptr += 4;
flags = extract_field(ptr, 16);
ptr += 2;
serverid = extract_field(ptr, 32);
ptr += 4;
strncpy(slave->binlogfile, (char *)ptr, binlognamelen);
slave->binlogfile[binlognamelen] = 0;
slave->seqno = 1;
if (slave->nocrc)
len = 19 + 8 + binlognamelen;
else
len = 19 + 8 + 4 + binlognamelen;
// Build a fake rotate event
resp = gwbuf_alloc(len + 5);
hdr.payload_len = len + 1;
hdr.seqno = slave->seqno++;
hdr.ok = 0;
hdr.timestamp = 0L;
hdr.event_type = ROTATE_EVENT;
hdr.serverid = router->masterid;
hdr.event_size = len;
hdr.next_pos = 0;
hdr.flags = 0x20;
ptr = blr_build_header(resp, &hdr);
encode_value(ptr, slave->binlog_pos, 64);
ptr += 8;
memcpy(ptr, slave->binlogfile, binlognamelen);
ptr += binlognamelen;
if (!slave->nocrc)
{
/*
* Now add the CRC to the fake binlog rotate event.
*
* The algorithm is first to compute the checksum of an empty buffer
* and then the checksum of the event portion of the message, ie we do not
* include the length, sequence number and ok byte that makes up the first
* 5 bytes of the message. We also do not include the 4 byte checksum itself.
*/
chksum = crc32(0L, NULL, 0);
chksum = crc32(chksum, GWBUF_DATA(resp) + 5, hdr.event_size - 4);
encode_value(ptr, chksum, 32);
}
rval = slave->dcb->func.write(slave->dcb, resp);
/* Send the FORMAT_DESCRIPTION_EVENT */
if (slave->binlog_pos != 4)
blr_slave_send_fde(router, slave);
slave->dcb->low_water = router->low_water;
slave->dcb->high_water = router->high_water;
dcb_add_callback(slave->dcb, DCB_REASON_DRAINED, blr_slave_callback, slave);
slave->state = BLRS_DUMPING;
LOGIF(LM, (skygw_log_write(
LOGFILE_MESSAGE,
"%s: New slave %s, server id %d, requested binlog file %s from position %lu",
router->service->name, slave->dcb->remote,
slave->serverid,
slave->binlogfile, slave->binlog_pos)));
if (slave->binlog_pos != router->binlog_position ||
strcmp(slave->binlogfile, router->binlog_name) != 0)
{
spinlock_acquire(&slave->catch_lock);
slave->cstate &= ~CS_UPTODATE;
slave->cstate |= CS_EXPECTCB;
spinlock_release(&slave->catch_lock);
poll_fake_write_event(slave->dcb);
}
return rval;
}
