// this host is declared dead; deal with the loss of data
//
int handle_host(DB_HOST& h) {
DB_VDA_CHUNK_HOST ch;
char buf[256];
int retval;
log_messages.printf(MSG_NORMAL, "processing dead host %d\n", h.id);
sprintf(buf, "where host_id=%d", h.id);
while (1) {
retval = ch.enumerate(buf);
if (retval == ERR_DB_NOT_FOUND) break;
if (retval) return retval;
log_messages.printf(MSG_NORMAL, " updating file%d\n", ch.vda_file_id);
DB_VDA_FILE vf;
retval = vf.lookup_id(ch.vda_file_id);
if (retval) {
log_messages.printf(MSG_CRITICAL,
" file lookup failed%d\n", ch.vda_file_id
);
return retval;
}
retval = vf.update_field("need_update=1");
if (retval) {
log_messages.printf(MSG_CRITICAL,
" file update failed%d\n", ch.vda_file_id
);
return retval;
}
}
return 0;
}
int CHUNK::start_upload() {
// if no upload of this chunk is in progress, start one.
// NOTE: all instances are inherently present_on_host,
// since this is only called if chunk is not present on server
//
VDA_CHUNK_HOST* chp;
set<VDA_CHUNK_HOST*>::iterator i;
for (i=hosts.begin(); i!=hosts.end(); i++) {
chp = *i;
if (chp->transfer_in_progress) return 0;
}
chp = *(hosts.begin());
DB_VDA_CHUNK_HOST dch;
char set_clause[256], where_clause[256];
sprintf(set_clause,
"transfer_in_progress=1, transfer_wait=1, transfer_request_time=%f",
dtime()
);
sprintf(where_clause,
"where vda_file_id=%d and host_id=%d and name='%s'",
chp->vda_file_id,
chp->host_id,
name
);
int retval = dch.update_fields_noid(set_clause, where_clause);
return retval;
}
int handle_remove(const char* name) {
DB_VDA_FILE vf;
char buf[1024];
sprintf(buf, "where file_name='%s'", name);
int retval = vf.lookup(buf);
if (retval) return retval;
// delete DB records
//
DB_VDA_CHUNK_HOST ch;
sprintf(buf, "vda_file_id=%d", vf.id);
ch.delete_from_db_multi(buf);
vf.delete_from_db();
// remove symlink from download hier
//
dir_hier_path(name, config.download_dir, config.uldl_dir_fanout, buf);
unlink(buf);
// remove encoded data and directories
//
retval = chdir(vf.dir);
if (retval) perror("chdir");
retval = system("/bin/rm -r [0-9]* Coding data.vda");
if (retval) perror("system");
return 0;
}
// Process a present file; possibilities:
// - a download finished
// - this host hasn't communicated in a while, and we deleted the
// VDA_CHUNK_HOST record
// So:
// - create a vda_chunk_host record if needed
// - set present_on_host flag in vda_chunk_host
// - mark our in-memory vda_chunk_host record as "found"
// - mark vda_file for update
//
static void process_present_file(FILE_INFO& fi, CHUNK_LIST& chunks) {
char fname[256], chunk_name[256], buf[256];
int hostid, retval;
retval = parse_physical_filename(fi.name, hostid, chunk_name, fname);
if (retval) {
log_messages.printf(MSG_CRITICAL,
"Can't parse VDA filename %s\n", fi.name
);
return;
}
DB_VDA_FILE vf;
sprintf(buf, "where file_name='%s'", fname);
retval = vf.lookup(buf);
if (retval) {
log_messages.printf(MSG_CRITICAL, "No VDA file %s\n", fname);
return;
}
if (fi.nbytes != vf.chunk_size) {
log_messages.printf(MSG_CRITICAL,
"wrong chunk size: %.0f != %.0f\n",
fi.nbytes, vf.chunk_size
);
return;
}
CHUNK_LIST::iterator cli = chunks.find(string(fi.name));
if (cli == chunks.end()) {
// don't have a record of this chunk on this host; make one
//
DB_VDA_CHUNK_HOST ch;
ch.create_time = dtime();
ch.vda_file_id = vf.id;
ch.host_id = g_reply->host.id;
strcpy(ch.physical_file_name, fi.name);
ch.present_on_host = true;
ch.transfer_in_progress = false;
ch.transfer_wait = false;
ch.transfer_request_time = 0;
ch.transfer_send_time = 0;
retval = ch.insert();
if (retval) {
log_messages.printf(MSG_CRITICAL, "ch.insert() failed\n");
return;
}
} else {
// update the existing record
//
DB_VDA_CHUNK_HOST* chp = &(cli->second);
chp->transfer_in_progress = false;
chp->transfer_wait = false;
chp->present_on_host = true;
chp->update();
}
mark_for_update(vf.id);
}
// get the state of an already-initialized file:
// expand the encoding tree,
// enumerate the VDA_HOST_CHUNKs from the DB
// and put them in the appropriate lists
//
int VDA_FILE_AUX::get_state() {
char buf[256];
sprintf(buf, "%s/chunk_sizes.txt", dir);
FILE* f = fopen(buf, "r");
if (!f) return -1;
for (int i=0; i<policy.coding_levels; i++) {
int n = fscanf(f, "%lf\n", &(policy.chunk_sizes[i]));
if (n != 1) {
fclose(f);
return -1;
}
}
fclose(f);
meta_chunk = new META_CHUNK(this, NULL, 0);
int retval = meta_chunk->get_state(dir, policy, 0);
if (retval) return retval;
DB_VDA_CHUNK_HOST vch;
sprintf(buf, "where vda_file_id=%d", id);
while (1) {
retval = vch.enumerate(buf);
if (retval == ERR_DB_NOT_FOUND) break;
if (retval) return retval;
vector<int> chunk_numbers;
retval = get_chunk_numbers(vch, chunk_numbers);
if (retval) {
log_messages.printf(MSG_CRITICAL,
"get_chunk_numbers(): %d\n", retval
);
return retval;
}
if ((int)(chunk_numbers.size()) != policy.coding_levels) {
log_messages.printf(MSG_CRITICAL,
"wrong get_chunk_numbers: got %d, expected %d\n",
(int)(chunk_numbers.size()), policy.coding_levels
);
return -1;
}
META_CHUNK* mc = meta_chunk;
for (int i=0; i<policy.coding_levels; i++) {
if (i == policy.coding_levels-1) {
CHUNK* c = (CHUNK*)(mc->children[chunk_numbers[i]]);
VDA_CHUNK_HOST* vchp = new VDA_CHUNK_HOST();
*vchp = vch;
c->hosts.insert(vchp);
} else {
mc = (META_CHUNK*)(mc->children[chunk_numbers[i]]);
}
}
}
return 0;
}
// assign this chunk to a host
//
int CHUNK::assign() {
int host_id = parent->dfile->choose_host();
if (!host_id) {
return ERR_NOT_FOUND;
}
DB_VDA_CHUNK_HOST ch;
ch.create_time = dtime();
ch.vda_file_id = parent->dfile->id;
ch.host_id = host_id;
strcpy(ch.name, name);
ch.size = parent->dfile->policy.chunk_size();
ch.present_on_host = 0;
ch.transfer_in_progress = true;
ch.transfer_wait = true;
ch.transfer_request_time = ch.create_time;
ch.transfer_send_time = 0;
int retval = ch.insert();
if (retval) {
log_messages.printf(MSG_CRITICAL, "ch.insert() failed\n");
return retval;
}
return 0;
}
// handle a scheduler request:
//
// - handle completed uploads
// - handle set of files present on client
// (update or create VDA_CHUNK_HOST record)
// - handle files expected but not present
// - issue delete commands if needed to enforce share
// - issue upload or download commands to client
//
// relevant fields of SCHEDULER_REQUEST
// file_infos: list of sticky files
// file_xfer_results: list of completed file xfers
//
void handle_vda() {
int retval;
unsigned int i;
CHUNK_LIST chunks;
// chunks that are supposed to be on this host
// if client is outdated, mark as dead
//
if (outdated_client(g_reply->host)) {
g_reply->host.cpu_efficiency = 1;
return;
}
// otherwise mark it as alive
//
g_reply->host.cpu_efficiency = 0;
// enumerate the vda_chunk_host records for this host from DB
//
DB_VDA_CHUNK_HOST ch;
char buf[256];
sprintf(buf, "where host_id=%d", g_reply->host.id);
while (1) {
retval = ch.enumerate(buf);
if (retval == ERR_DB_NOT_FOUND) break;
if (retval) {
// if we didn't get a complete enumeration,
// give up rather than continuing with partial info
//
log_messages.printf(MSG_NORMAL,
"[vda]: ch.enumerate() failed %d\n", retval
);
return;
}
if (config.debug_vda) {
log_messages.printf(MSG_NORMAL,
"[vda] DB: has chunk %s, file %d\n",
ch.physical_file_name, ch.vda_file_id
);
}
chunks.insert(
pair<string, DB_VDA_CHUNK_HOST>(string(ch.physical_file_name), ch)
);
}
// process completed uploads
// (completed downloads are handled below)
//
for (i=0; i<g_request->file_xfer_results.size(); i++) {
RESULT& r = g_request->file_xfer_results[i];
if (strstr(r.name, "upload_vda_")) {
char* phys_file_name = r.name + strlen("upload_");
if (config.debug_vda) {
log_messages.printf(MSG_NORMAL,
"[vda] completed upload of %s\n", phys_file_name
);
}
retval = process_completed_upload(phys_file_name, chunks);
if (retval) {
log_messages.printf(MSG_CRITICAL,
"[vda] process_completed_upload(): %d\n", retval
);
return;
}
}
}
// process files present on client
//
for (i=0; i<g_request->file_infos.size(); i++) {
FILE_INFO& fi = g_request->file_infos[i];
if (!starts_with(fi.name, "vda_")) {
continue;
}
if (config.debug_vda) {
log_messages.printf(MSG_NORMAL,
"[vda] request: client has file %s, status %d\n",
fi.name, fi.status
);
}
if (fi.status != FILE_PRESENT) {
continue;
}
process_chunk_present_on_client(fi, chunks);
}
process_chunks_missing_on_client(chunks);
enforce_quota(chunks);
issue_transfer_commands(chunks);
}
// Process a file that's present on client.
// A VDA_CHUNK_HOST record may not be in the DB,
// e.g. because this host hasn't communicated in a while
// and we deleted the VDA_CHUNK_HOST record
// So:
// - create a vda_chunk_host record if needed
// - set present_on_host flag in vda_chunk_host
// - mark our in-memory vda_chunk_host record as "found"
// - mark vda_file for update
//
static void process_chunk_present_on_client(FILE_INFO& fi, CHUNK_LIST& chunks) {
char fname[256], chunk_name[256], buf[1024];
int hostid, retval;
retval = parse_physical_filename(fi.name, hostid, chunk_name, fname);
if (retval) {
log_messages.printf(MSG_CRITICAL,
"Can't parse VDA filename %s\n", fi.name
);
return;
}
DB_VDA_FILE vf;
sprintf(buf, "where file_name='%s'", fname);
retval = vf.lookup(buf);
if (retval) {
log_messages.printf(MSG_CRITICAL,
"No VDA file for %s, deleting\n", fi.name
);
delete_file_xml(fi.name, buf);
g_reply->file_transfer_requests.push_back(string(buf));
return;
}
if (fi.nbytes != vf.chunk_size) {
log_messages.printf(MSG_CRITICAL,
"wrong chunk size for %s: %.0f != %.0f, deleting\n",
fi.name, fi.nbytes, vf.chunk_size
);
delete_file_xml(fi.name, buf);
g_reply->file_transfer_requests.push_back(string(buf));
return;
}
CHUNK_LIST::iterator cli = chunks.find(string(fi.name));
if (cli == chunks.end()) {
// we don't have a record of this chunk on this host; make one
//
DB_VDA_CHUNK_HOST ch;
ch.create_time = dtime();
ch.vda_file_id = vf.id;
ch.host_id = g_reply->host.id;
strcpy(ch.physical_file_name, fi.name);
ch.present_on_host = true;
ch.transfer_in_progress = false;
ch.transfer_wait = false;
ch.transfer_request_time = 0;
ch.transfer_send_time = 0;
retval = ch.insert();
if (retval) {
log_messages.printf(MSG_CRITICAL, "ch.insert() failed\n");
return;
}
mark_for_update(vf.id);
} else {
// we already have a DB record.
// If needed, update it and mark file for update
//
DB_VDA_CHUNK_HOST* chp = &(cli->second);
chp->found = true;
// if file wasn't previously on host, update the main file
//
if (!chp->present_on_host) {
mark_for_update(vf.id);
chp->transfer_in_progress = false;
chp->transfer_wait = false;
chp->present_on_host = true;
sprintf(buf,
"host_id=%d and physical_file_name='%s'",
chp->host_id, chp->physical_file_name
);
chp->update_fields_noid(
"transfer_in_progress=0, transfer_wait=0, present_on_host=1",
buf
);
}
}
}
// Pick a host to send a chunk of this file to.
// The host must:
// 1) be alive (recent RPC time)
// 2) not have any chunks of this file
//
// We maintain a cache of such hosts
// The policy is:
//
// - scan the cache, removing hosts that are no longer alive;
// return if find a live host
// - pick a random starting point in host ID space,
// and enumerate 100 live hosts; wrap around if needed.
// Return one and put the rest in cache
//
int VDA_FILE_AUX::choose_host() {
int retval;
DB_HOST host;
return 467;
// replenish cache if needed
//
if (!available_hosts.size()) {
int nhosts_scanned = 0;
int rand_id;
for (int i=0; i<2; i++) {
char buf[256];
if (i == 0) {
retval = host.max_id(rand_id, "");
if (retval) {
log_messages.printf(MSG_CRITICAL, "host.max_id() failed\n");
return 0;
}
rand_id = (int)(((double)id)*drand());
sprintf(buf,
"where %s and id>=%d order by id limit 100",
host_alive_clause(), rand_id
);
} else {
sprintf(buf,
"where %s and id<%d order by id limit %d",
host_alive_clause(), rand_id, 100-nhosts_scanned
);
}
while (1) {
retval = host.enumerate(buf);
if (retval == ERR_DB_NOT_FOUND) break;
if (retval) {
log_messages.printf(MSG_CRITICAL, "host enum failed\n");
return 0;
}
nhosts_scanned++;
DB_VDA_CHUNK_HOST ch;
char buf2[256];
int count;
sprintf(buf2, "where vda_file_id=%d and host_id=%d", id, host.id);
#if 0
retval = ch.count(count, buf2);
if (retval) {
log_messages.printf(MSG_CRITICAL, "ch.count failed\n");
return 0;
}
#else
count = 0;
#endif
if (count == 0) {
available_hosts.push_back(host.id);
}
if (nhosts_scanned == 100) break;
}
if (nhosts_scanned == 100) break;
}
}
while (available_hosts.size()) {
int hostid = available_hosts.back();
available_hosts.pop_back();
retval = host.lookup_id(hostid);
if (retval || !alive(host)) {
continue;
}
return hostid;
}
log_messages.printf(MSG_CRITICAL, "No hosts available\n");
return 0;
}
// handle a scheduler request:
//
// - handle completed uploads
// - handle set of files present on client
// (update or create VDA_CHUNK_HOST record)
// - handle files expected but not present
// - issue delete commands if needed to enforce share
// - issue upload or download commands to client
//
// relevant fields of SCHEDULER_REQUEST
// file_infos: list of sticky files
// file_xfer_results: list of completed file xfers
//
void handle_vda() {
int retval;
unsigned int i;
CHUNK_LIST chunks;
// chunks that are supposed to be on this host
// enumerate the vda_chunk_host records for this host from DB
//
DB_VDA_CHUNK_HOST ch;
char buf[256];
sprintf(buf, "where host_id=%d", g_reply->host.id);
while (1) {
retval = ch.enumerate(buf);
if (retval == ERR_DB_NOT_FOUND) break;
if (retval) {
// if we didn't get a complete enumeration,
// give up rather than continuing with partial info
//
log_messages.printf(MSG_NORMAL,
"[vda]: ch.enumerate() failed %d\n", retval
);
return;
}
if (config.debug_vda) {
log_messages.printf(MSG_NORMAL,
"[vda] DB: has chunk %s\n", ch.name
);
}
chunks.insert(pair<string, DB_VDA_CHUNK_HOST>(string(ch.name), ch));
}
// process completed uploads
//
for (i=0; i<g_request->file_xfer_results.size(); i++) {
RESULT& r = g_request->file_xfer_results[i];
if (!starts_with(r.name, "vda_upload_")) continue;
char* chunk_name = r.name + strlen("vda_upload_");
if (config.debug_vda) {
log_messages.printf(MSG_NORMAL,
"[vda] DB: completed upload %s\n", chunk_name
);
}
retval = process_completed_upload(chunk_name, chunks);
if (retval) {
log_messages.printf(MSG_CRITICAL,
"[vda] process_completed_upload(): %d\n", retval
);
return;
}
}
// process files present on host
//
for (i=0; i<g_request->file_infos.size(); i++) {
FILE_INFO& fi = g_request->file_infos[i];
if (!starts_with(fi.name, "vda_")) {
continue;
}
if (config.debug_vda) {
log_messages.printf(MSG_NORMAL,
"[vda] request: client has file %s\n", fi.name
);
}
process_present_file(fi, chunks);
}
process_missing_chunks(chunks);
enforce_quota(chunks);
issue_transfer_commands(chunks);
}