static GError*
_resolve_reference_service(struct hc_resolver_s *r, struct hashstr_s *hk,
struct hc_url_s *u, const gchar *s, gchar ***result)
{
GError *err;
gchar **m1urlv = NULL;
GRID_TRACE2("%s(%s,%s,%s)", __FUNCTION__, hashstr_str(hk),
hc_url_get(u, HCURL_WHOLE), s);
/* Try to hit the cache for the service itself */
*result = hc_resolver_get_cached(r, r->services.cache, hk);
if (NULL != *result) {
return NULL;
}
/* now attempt a real resolution */
err = _resolve_meta1(r, u, &m1urlv);
g_assert((err!=NULL) ^ (m1urlv!=NULL));
if (NULL != err)
return err;
err = _resolve_service_through_many_meta1(m1urlv, u, s, result);
g_assert((err!=NULL) ^ (*result!=NULL));
if (!err) {
/* fill the cache */
if (!(r->flags & HC_RESOLVER_NOCACHE))
hc_resolver_store(r, r->services.cache, hk, *result);
}
g_strfreev(m1urlv);
return err;
}
static GError *
_resolve_m1_through_many_m0(gchar **urlv, const guint8 *prefix, gchar ***result)
{
GError *err;
guint i, last;
gchar *url;
GRID_TRACE2("%s(%02X%02X)", __FUNCTION__, prefix[0], prefix[1]);
for (last=g_strv_length(urlv); last ;last--) {
/* pick a random URL */
i = rand() % last;
url = urlv[i];
if (!(err = _resolve_m1_through_one_m0(url, prefix, result)))
return NULL;
if (err->code < 100)
g_error_free(err);
/* swap 'i' and 'last' */
urlv[i] = urlv[last-1];
urlv[last-1] = url;
}
return NEWERROR(500, "No META0 answered");
}
GError *
sqlx_cache_unlock_and_close_base(sqlx_cache_t *cache, gint bd, gboolean force)
{
GError *err = NULL;
GRID_TRACE2("%s(%p,%d,%d)", __FUNCTION__, (void*)cache, bd, force);
EXTRA_ASSERT(cache != NULL);
if (base_id_out(cache, bd))
return NEWERROR(CODE_INTERNAL_ERROR, "invalid base id=%d", bd);
g_mutex_lock(&cache->lock);
cache->used = TRUE;
sqlx_base_t *base; base = GET(cache,bd);
switch (base->status) {
case SQLX_BASE_FREE:
EXTRA_ASSERT(base->count_open == 0);
EXTRA_ASSERT(base->owner == NULL);
err = NEWERROR(CODE_INTERNAL_ERROR, "base not used");
break;
case SQLX_BASE_IDLE:
case SQLX_BASE_IDLE_HOT:
EXTRA_ASSERT(base->count_open == 0);
EXTRA_ASSERT(base->owner == NULL);
err = NEWERROR(CODE_INTERNAL_ERROR, "base closed");
break;
case SQLX_BASE_USED:
EXTRA_ASSERT(base->count_open > 0);
// held by the current thread
if (!(-- base->count_open)) { // to be closed
if (force) {
_expire_base(cache, base);
} else {
sqlx_base_debug("CLOSING", base);
base->owner = NULL;
if (base->heat >= cache->heat_threshold)
sqlx_base_move_to_list(cache, base, SQLX_BASE_IDLE_HOT);
else
sqlx_base_move_to_list(cache, base, SQLX_BASE_IDLE);
}
}
break;
case SQLX_BASE_CLOSING:
EXTRA_ASSERT(base->owner != NULL);
EXTRA_ASSERT(base->owner != g_thread_self());
err = NEWERROR(CODE_INTERNAL_ERROR, "base being closed");
break;
}
if (base && !err)
sqlx_base_debug(__FUNCTION__, base);
g_cond_signal(base->cond);
g_mutex_unlock(&cache->lock);
return err;
}
static GError *
_resolve_meta1(struct hc_resolver_s *r, struct hc_url_s *u, gchar ***result)
{
struct hashstr_s *hk;
GError *err = NULL;
GRID_TRACE2("%s(%s)", __FUNCTION__, hc_url_get(u, HCURL_WHOLE));
hk = hashstr_printf("meta1|%s|%.4s",
hc_url_get(u, HCURL_NSPHYS),
hc_url_get(u, HCURL_HEXID));
/* Try to hit the cache */
if (!(*result = hc_resolver_get_cached(r, r->csm0.cache, hk))) {
/* get a meta0, then store it in the cache */
gchar **m0urlv = NULL;
err = _resolve_meta0(r, hc_url_get(u, HCURL_NSPHYS), &m0urlv);
if (err != NULL)
g_prefix_error(&err, "M0 resolution error: ");
else {
err = _resolve_m1_through_many_m0(m0urlv, hc_url_get_id(u), result);
if (!err)
hc_resolver_store(r, r->csm0.cache, hk, *result);
g_strfreev(m0urlv);
}
}
g_free(hk);
return err;
}
static GError*
_resolve_meta0(struct hc_resolver_s *r, const gchar *ns, gchar ***result)
{
struct hashstr_s *hk;
GError *err = NULL;
GRID_TRACE2("%s(%s)", __FUNCTION__, ns);
hk = hashstr_printf("meta0|%s", ns);
/* Try to hit the cache */
if (!(*result = hc_resolver_get_cached(r, r->csm0.cache, hk))) {
GSList *allm0;
/* Now attempt a real resolution */
if (!(allm0 = list_namespace_services2(ns, "meta0", &err))) {
if (!err)
err = NEWERROR(500, "No meta0 available");
*result = NULL;
}
else {
*result = _srvlit_to_urlv(allm0);
g_slist_foreach(allm0, service_info_gclean, NULL);
g_slist_free(allm0);
allm0 = NULL;
/* then fill the cache */
hc_resolver_store(r, r->csm0.cache, hk, *result);
err = NULL;
}
}
g_free(hk);
return err;
}
GError*
hc_resolve_reference_directory(struct hc_resolver_s *r, struct hc_url_s *url,
gchar ***result)
{
GRID_TRACE2("%s(%s)", __FUNCTION__, hc_url_get(url, HCURL_WHOLE));
EXTRA_ASSERT(r != NULL);
EXTRA_ASSERT(url != NULL);
EXTRA_ASSERT(result != NULL);
if (!hc_url_get_id(url) || !hc_url_has(url, HCURL_NS))
return NEWERROR(CODE_BAD_REQUEST, "Incomplete URL [%s]", hc_url_get(url, HCURL_WHOLE));
GError *err = NULL;
gchar **m1v = NULL, **m0v = NULL;
if (!(err = _resolve_meta0(r, hc_url_get(url, HCURL_NS), &m0v)))
err = _resolve_meta1(r, url, &m1v);
if (err) {
if (m0v) g_strfreev (m0v);
if (m1v) g_strfreev (m1v);
return err;
}
*result = g_malloc0(sizeof(gchar*) *
(g_strv_length(m0v) + g_strv_length(m1v) + 1));
gchar **d = *result;
for (gchar **p=m0v; *p ;++p) { *(d++) = *p; }
g_free (m0v); // pointers reused
for (gchar **p=m1v; *p ;++p) { *(d++) = *p; }
g_free (m1v); // pointers reused
return NULL;
}
static GError *
_resolve_m1_through_many_m0(gchar **urlv, const guint8 *prefix, gchar ***result)
{
guint i, last;
gchar *url;
GRID_TRACE2("%s(%02X%02X)", __FUNCTION__, prefix[0], prefix[1]);
for (last=g_strv_length(urlv); last ;last--) {
/* pick a random URL */
i = rand() % last;
url = urlv[i];
GError *err = _resolve_m1_through_one_m0(url, prefix, result);
EXTRA_ASSERT((err!=NULL) ^ (*result!=NULL));
if (!err)
return NULL;
if (!CODE_IS_NETWORK_ERROR(err->code))
return err;
g_error_free(err);
/* swap 'i' and 'last' */
urlv[i] = urlv[last-1];
urlv[last-1] = url;
}
return NEWERROR(CODE_INTERNAL_ERROR, "No META0 answered");
}
GError*
hc_resolve_reference_service(struct hc_resolver_s *r, struct hc_url_s *url,
const gchar *srvtype, gchar ***result)
{
GError *err;
struct hashstr_s *hk;
GRID_TRACE2("%s(%s,%s)", __FUNCTION__, hc_url_get(url, HCURL_WHOLE), srvtype);
g_assert(r != NULL);
g_assert(url != NULL);
g_assert(srvtype != NULL);
g_assert(result != NULL);
g_assert(*result == NULL);
if (!hc_url_get_id(url) || !hc_url_has(url, HCURL_NS))
return NEWERROR(400, "Incomplete URL [%s]", hc_url_get(url, HCURL_WHOLE));
hk = hashstr_printf("%s|%s|%s", srvtype,
hc_url_get(url, HCURL_NSPHYS),
hc_url_get(url, HCURL_HEXID));
err = _resolve_reference_service(r, hk, url, srvtype, result);
g_free(hk);
return err;
}
static GError *
_resolve_service_through_many_meta1(gchar **urlv, struct hc_url_s *u,
const gchar *s, gchar ***result)
{
guint i, last;
gchar *url;
GRID_TRACE2("%s(%s,%s)", __FUNCTION__, hc_url_get(u, HCURL_WHOLE), s);
for (last=g_strv_length(urlv); last ;last--) {
/* pick a random URL */
i = rand() % last;
url = urlv[i];
GError *err = _resolve_service_through_one_m1(url, u, s, result);
g_assert((err!=NULL) ^ (*result!=NULL));
if (!err)
return NULL;
if (err->code >= 100)
return err;
g_clear_error(&err);
/* swap 'i' and 'last' */
urlv[i] = urlv[last-1];
urlv[last-1] = url;
}
return NEWERROR(500, "No META0 answered");
}
static GError *
_resolve_m1_through_one_m0(const gchar *m0, const guint8 *prefix, gchar ***result)
{
GError *err = NULL;
struct addr_info_s ai;
GRID_TRACE2("%s(%s,%02X%02X)", __FUNCTION__, m0, prefix[0], prefix[1]);
meta1_strurl_get_address(m0, &ai);
do {
GSList *lmap = meta0_remote_get_meta1_one(&ai,
_timeout(&rc_resolver_timeout_m0, 30.0), prefix, &err);
if (!lmap) {
if (err)
return err;
return NEWERROR(500, "No meta1 found");
}
else {
*result = _m0list_to_urlv(lmap);
g_slist_foreach(lmap, meta0_info_gclean, NULL);
g_slist_free(lmap);
err = NULL;
}
} while (0);
return err;
}
static enum http_rc_e
handler_action (struct http_request_s *rq, struct http_reply_ctx_s *rp)
{
// Get a request id for the current request
const gchar *reqid = g_tree_lookup (rq->tree_headers, PROXYD_HEADER_REQID);
if (reqid)
oio_ext_set_reqid(reqid);
else
oio_ext_set_random_reqid();
// Then parse the request to find a handler
struct oio_url_s *url = NULL;
struct oio_requri_s ruri = {NULL, NULL, NULL, NULL};
oio_requri_parse (rq->req_uri, &ruri);
struct path_matching_s **matchings = _metacd_match (rq->cmd, ruri.path);
GRID_TRACE2("URI path[%s] query[%s] fragment[%s] matches[%u]",
ruri.path, ruri.query, ruri.fragment,
g_strv_length((gchar**)matchings));
GQuark gq_count = gq_count_unexpected;
GQuark gq_time = gq_time_unexpected;
enum http_rc_e rc;
if (!*matchings) {
rp->set_content_type ("application/json");
rp->set_body_gstr (g_string_new("{\"status\":404,\"message\":\"No handler found\"}"));
rp->set_status (HTTP_CODE_NOT_FOUND, "No handler found");
rp->finalize ();
rc = HTTPRC_DONE;
} else {
struct req_args_s args = {0};
args.req_uri = &ruri;
args.matchings = matchings;
args.rq = rq;
args.rp = rp;
args.url = url = _metacd_load_url (&args);
rp->subject(oio_url_get(url, OIOURL_HEXID));
gq_count = (*matchings)->last->gq_count;
gq_time = (*matchings)->last->gq_time;
GRID_TRACE("%s %s URL %s", __FUNCTION__, ruri.path, oio_url_get(args.url, OIOURL_WHOLE));
req_handler_f handler = (*matchings)->last->u;
rc = (*handler) (&args);
}
gint64 spent = oio_ext_monotonic_time () - rq->client->time.evt_in;
network_server_stat_push4 (rq->client->server, TRUE,
gq_count, 1, gq_count_all, 1,
gq_time, spent, gq_time_all, spent);
path_matching_cleanv (matchings);
oio_requri_clear (&ruri);
oio_url_pclean (&url);
oio_ext_set_reqid (NULL);
return rc;
}
void
sqlx_cache_set_max_bases(sqlx_cache_t *cache, guint max)
{
GRID_TRACE2("%s(%p,%u)", __FUNCTION__, cache, max);
EXTRA_ASSERT(cache != NULL);
EXTRA_ASSERT(max < 65536);
sqlx_cache_reset_bases(cache, max);
}
static gpointer
_row_to_bean(const struct bean_descriptor_s *descr, sqlite3_stmt *stmt)
{
const struct field_descriptor_s *fd;
gpointer res;
int col, s;
res = _bean_create(descr);
for (fd=descr->fields; fd->name ;fd++) {
col = fd->position;
#if 0
GRID_TRACE2("SQL column[%d,%d/%s,%s] field[%d,%ld,%d/%s,%s]",
col,
sqlite3_column_type(stmt, col),
sqlite_strtype(sqlite3_column_type(stmt, col)),
sqlite3_column_name(stmt, col),
fd->position,
fd->offset,
fd->type,
bean_strtype(fd->type),
fd->name);
#endif
if (sqlite3_column_type(stmt, col) == SQLITE_NULL) {
_bean_del_field(res, fd->position);
continue;
}
_bean_set_field(res, fd->position);
gpointer pf = ((guint8*)res) + descr->offset_fields + fd->offset;
switch (fd->type) {
case FT_BOOL:
*((gboolean*)pf) = sqlite3_column_int(stmt, col);
continue;
case FT_INT:
*((gint64*)pf) = sqlite3_column_int64(stmt, col);
continue;
case FT_REAL:
*((gdouble*)pf) = sqlite3_column_double(stmt, col);
continue;
case FT_TEXT:
s = sqlite3_column_bytes(stmt, col);
g_string_append_len(GSTR(pf), (const gchar*)sqlite3_column_text(stmt, col), s);
continue;
case FT_BLOB:
s = sqlite3_column_bytes(stmt, col);
g_byte_array_append(GBA(pf), (guint8*)sqlite3_column_blob(stmt, col), s);
continue;
default:
g_assert_not_reached();
continue;
}
}
return res;
}
GError*
hc_resolve_reference_directory(struct hc_resolver_s *r, struct hc_url_s *url,
gchar ***result)
{
GRID_TRACE2("%s(%s)", __FUNCTION__, hc_url_get(url, HCURL_WHOLE));
g_assert(r != NULL);
g_assert(url != NULL);
g_assert(result != NULL);
if (!hc_url_get_id(url) || !hc_url_has(url, HCURL_NS))
return NEWERROR(400, "Incomplete URL [%s]", hc_url_get(url, HCURL_WHOLE));
return _resolve_meta1(r, url, result);
}
static GError *
_resolve_service_through_one_m1(const gchar *m1, struct hc_url_s *u,
const gchar *s, gchar ***result)
{
GError *err = NULL;
struct addr_info_s ai;
GRID_TRACE2("%s(%s,%s,%s)", __FUNCTION__, m1, hc_url_get(u, HCURL_WHOLE), s);
meta1_strurl_get_address(m1, &ai);
*result = meta1v2_remote_list_reference_services(&ai, &err, u, s);
EXTRA_ASSERT((err!=NULL) ^ (*result!=NULL));
return err;
}
static void
sqlx_base_move_to_list(sqlx_cache_t *cache, sqlx_base_t *base,
enum sqlx_base_status_e status)
{
register enum sqlx_base_status_e status0;
if (status != (status0 = base->status)) {
sqlx_base_remove_from_list(cache, base);
sqlx_base_add_to_list(cache, base, status);
}
GRID_TRACE2("BASE [%d/%s] moved from %s to %s",
base->index,
hashstr_str(base->name),
sqlx_status_to_str(status0),
sqlx_status_to_str(status));
}
static void
sqlx_base_debug_func(const gchar *from, sqlx_base_t *base)
{
(void) from;
(void) base;
EXTRA_ASSERT(base);
GRID_TRACE2("BASE [%d/%s]"
" %"G_GUINT32_FORMAT
" LIST=%s [%d,%d]"
" (%s)",
base->index, (base->name ? hashstr_str(base->name) : ""),
base->count_open,
sqlx_status_to_str(base->status),
base->link.prev, base->link.next,
from);
}
void
hc_decache_reference_service(struct hc_resolver_s *r, struct hc_url_s *url,
const gchar *srvtype)
{
struct hashstr_s *hk;
GRID_TRACE2("%s(%s,%s)", __FUNCTION__, hc_url_get(url, HCURL_WHOLE), srvtype);
EXTRA_ASSERT(r != NULL);
EXTRA_ASSERT(url != NULL);
EXTRA_ASSERT(srvtype != NULL);
if (r->flags & HC_RESOLVER_NOCACHE)
return;
hk = hashstr_printf("%s|%s|%s", srvtype,
hc_url_get(url, HCURL_NS), hc_url_get(url, HCURL_HEXID));
hc_resolver_forget(r, r->services.cache, hk);
g_free(hk);
}
static void
load_table_row(sqlite3 *db, const hashstr_t *name, gint64 rowid, Row_t *row,
Table_t *table)
{
int rc;
sqlite3_stmt *stmt = NULL;
gchar *sql;
GRID_TRACE2("%s(%p,%s,%"G_GINT64_FORMAT",%p,%p)", __FUNCTION__,
db, hashstr_str(name), rowid, row, table);
sql = g_strdup_printf("SELECT * FROM %s WHERE ROWID = ?", hashstr_str(name));
sqlite3_prepare_debug(rc, db, sql, -1, &stmt, NULL);
g_free(sql);
sqlite3_bind_int64(stmt, 1, rowid);
while (SQLITE_ROW == (rc = sqlite3_step(stmt)))
load_statement(stmt, row, table);
sqlite3_finalize_debug(rc, stmt);
}
void
hc_decache_reference(struct hc_resolver_s *r, struct hc_url_s *url)
{
struct hashstr_s *hk;
GRID_TRACE2("%s(%s)", __FUNCTION__, hc_url_get(url, HCURL_WHOLE));
g_assert(r != NULL);
g_assert(url != NULL);
if (r->flags & HC_RESOLVER_NOCACHE)
return;
hk = hashstr_printf("meta0|%s", hc_url_get(url, HCURL_NSPHYS));
hc_resolver_forget(r, r->csm0.cache, hk);
g_free(hk);
hk = hashstr_printf("meta1|%s|%.4s", hc_url_get(url, HCURL_NSPHYS),
hc_url_get(url, HCURL_HEXID));
hc_resolver_forget(r, r->csm0.cache, hk);
g_free(hk);
}
static GError*
_load(struct meta0_backend_s *m0)
{
GError *err = NULL;
struct sqlx_sqlite3_s *sq3 = NULL;
GRID_TRACE2("%s(%p)", __FUNCTION__, m0);
err = _open_and_lock(m0,M0V2_OPENBASE_MASTERSLAVE, &sq3);
if (err != NULL) {
return err;
}
err = _load_from_base(sq3, &(m0->array_by_prefix));
if (err != NULL)
g_prefix_error(&err, "Query error: ");
err = _load_meta1ref_from_base(sq3, &(m0->array_meta1_ref));
if (err != NULL)
g_prefix_error(&err, "Query error: ");
_unlock_and_close(sq3);
return err;
}
GError *
sqlx_cache_open_and_lock_base(sqlx_cache_t *cache, const hashstr_t *hname,
gint *result)
{
gint bd;
GError *err = NULL;
sqlx_base_t *base = NULL;
GTimeVal *deadline = g_alloca(sizeof(GTimeVal));
GRID_TRACE2("%s(%p,%s,%p)", __FUNCTION__, (void*)cache,
hname ? hashstr_str(hname) : "NULL", (void*)result);
EXTRA_ASSERT(cache != NULL);
EXTRA_ASSERT(hname != NULL);
EXTRA_ASSERT(result != NULL);
if (cache->open_timeout >= 0) {
g_get_current_time(deadline);
g_time_val_add(deadline, cache->open_timeout * 1000);
} else {
// wait forever
deadline = NULL;
}
g_mutex_lock(cache->lock);
cache->used = TRUE;
retry:
bd = sqlx_lookup_id(cache, hname);
if (bd < 0) {
if (!(err = sqlx_base_reserve(cache, hname, &base))) {
bd = base->index;
*result = base->index;
sqlx_base_debug("OPEN", base);
}
else {
GRID_DEBUG("No base available for [%s] (%d %s)",
hashstr_str(hname), err->code, err->message);
if (sqlx_expire_first_idle_base(cache, NULL) >= 0) {
g_clear_error(&err);
goto retry;
}
}
}
else {
base = GET(cache, bd);
switch (base->status) {
case SQLX_BASE_FREE:
EXTRA_ASSERT(base->count_open == 0);
EXTRA_ASSERT(base->owner == NULL);
GRID_ERROR("free base referenced");
g_assert_not_reached();
break;
case SQLX_BASE_IDLE:
case SQLX_BASE_IDLE_HOT:
EXTRA_ASSERT(base->count_open == 0);
EXTRA_ASSERT(base->owner == NULL);
sqlx_base_move_to_list(cache, base, SQLX_BASE_USED);
base->count_open ++;
base->owner = g_thread_self();
*result = base->index;
break;
case SQLX_BASE_USED:
EXTRA_ASSERT(base->count_open > 0);
EXTRA_ASSERT(base->owner != NULL);
if (base->owner != g_thread_self()) {
// The lock is held by another thread/request
GRID_DEBUG("Base [%s] in use by another thread (%X), waiting...",
hashstr_str(hname), compute_thread_id(base->owner));
/* This is to avoid server thread starvation,
* due to all threads waiting on the same base. */
if (cache->max_waiting > 0 &&
base->count_waiting >= cache->max_waiting) {
err = NEWERROR(CODE_UNAVAILABLE,
"database currently in use by another request, "
"and %d others threads already waiting",
base->count_waiting);
break;
}
base->count_waiting++;
if (g_cond_timed_wait(base->cond, cache->lock, deadline)) {
// Thread was woken up before deadline
GRID_DEBUG("Retrying to open [%s]", hashstr_str(hname));
base->count_waiting--;
goto retry;
} else {
// Deadline has been reached
base->count_waiting--;
if (cache->open_timeout > 0) {
err = NEWERROR(CODE_UNAVAILABLE,
"database currently in use by another request"
" (we waited %ldms)",
cache->open_timeout);
} else {
err = NEWERROR(CODE_UNAVAILABLE,
"database currently in use by another request");
}
GRID_DEBUG("failed to open base: "
"in use by another request (thread %X)",
compute_thread_id(base->owner));
break;
}
}
base->owner = g_thread_self();
base->count_open ++;
*result = base->index;
break;
case SQLX_BASE_CLOSING:
EXTRA_ASSERT(base->owner != NULL);
// Just wait for a notification then retry
if (g_cond_timed_wait(base->cond, cache->lock, deadline))
goto retry;
else {
err = NEWERROR(CODE_UNAVAILABLE,
"Database stuck in closing state");
break;
}
}
}
if (base) {
if (!err) {
sqlx_base_debug(__FUNCTION__, base);
EXTRA_ASSERT(base->owner == g_thread_self());
EXTRA_ASSERT(base->count_open > 0);
}
g_cond_signal(base->cond);
}
g_mutex_unlock(cache->lock);
return err;
}
static int
_client_manage_input(struct network_client_s *client)
{
guint total, size;
int _notify(void) {
if (!client->transport.notify_input)
return RC_PROCESSED;
if (!data_slab_sequence_has_data(&(client->input))) {
/* drain the data */
data_slab_sequence_clean_data(&(client->input));
return RC_PROCESSED;
}
GRID_TRACE2("fd=%d passing %u/%"G_GSIZE_FORMAT" to transport %p",
client->fd, total,
data_slab_sequence_size(&(client->input)),
client->transport.notify_input);
return client->transport.notify_input(client);
}
EXTRA_ASSERT(client != NULL);
EXTRA_ASSERT(client->fd >= 0);
for (size=SLAB_STARTSIZE, total=0; total < ROUND_MAXSIZE ;) {
int rc;
struct data_slab_s *in = data_slab_make_empty(size);
switch (rc = _ds_feed(client->fd, in)) {
case RC_ERROR:
data_slab_free(in);
return RC_ERROR;
case RC_NODATA: /* no more data to expect */
case RC_NOTREADY:
if (!in->data.buffer.end)
data_slab_free(in);
else {
data_slab_sequence_append(&(client->input), in);
total += in->data.buffer.end;
}
in = NULL;
if (RC_NODATA == _notify())
rc = RC_NODATA;
return rc;
case RC_PROCESSED:
if (!in->data.buffer.end)
data_slab_free(in);
else {
data_slab_sequence_append(&(client->input), in);
total += in->data.buffer.end;
}
size = SLAB_MAXSIZE;
in = NULL;
break;
default:
g_assert_not_reached();
}
}
return _notify();
}
struct conscience_srv_s *
conscience_srvtype_refresh(struct conscience_srvtype_s *srvtype, struct service_info_s *si)
{
g_assert_nonnull (srvtype);
g_assert_nonnull (si);
struct conscience_srvid_s srvid;
memcpy(&(srvid.addr), &(si->addr), sizeof(addr_info_t));
struct service_tag_s *tag_first =
service_info_get_tag(si->tags, NAME_TAGNAME_RAWX_FIRST);
gboolean really_first = FALSE;
/*register the service if necessary */
struct conscience_srv_s *p_srv = conscience_srvtype_get_srv(srvtype, &srvid);
if (!p_srv) {
p_srv = conscience_srvtype_register_srv(srvtype, NULL, &srvid);
g_assert_nonnull (p_srv);
really_first = tag_first && tag_first->type == STVT_BOOL && tag_first->value.b;
}
/* refresh the tags: create missing, replace existing
* (but the tags are not flushed before) */
if (si->tags) {
TRACE("Refreshing tags for srv [%.*s]",
(int)(LIMIT_LENGTH_SRVDESCR), p_srv->description);
const guint max = si->tags->len;
for (guint i = 0; i < max; i++) {
struct service_tag_s *tag = g_ptr_array_index(si->tags, i);
if (tag == tag_first) continue;
struct service_tag_s *orig = conscience_srv_ensure_tag(p_srv, tag->name);
service_tag_copy(orig, tag);
}
}
p_srv->score.timestamp = oio_ext_monotonic_seconds ();
if (si->score.value == SCORE_UNSET || si->score.value == SCORE_UNLOCK) {
if (really_first) {
GRID_TRACE2("SRV first [%s]", p_srv->description);
p_srv->score.value = 0;
p_srv->locked = TRUE;
} else {
if (si->score.value == SCORE_UNLOCK) {
if (p_srv->locked) {
GRID_TRACE2("SRV unlocked [%s]", p_srv->description);
p_srv->locked = FALSE;
p_srv->score.value = CLAMP (p_srv->score.value, SCORE_DOWN, SCORE_MAX);
} else {
GRID_TRACE2("SRV already unlocked [%s]", p_srv->description);
}
} else { /* UNSET, a.k.a. regular computation */
if (p_srv->locked) {
GRID_TRACE2("SRV untouched [%s]", p_srv->description);
} else {
GError *err = NULL;
if (!conscience_srv_compute_score(p_srv, &err)) {
GRID_TRACE2("SRV error [%s]: (%d) %s", p_srv->description, err->code, err->message);
g_clear_error (&err);
} else {
GRID_TRACE2("SRV refreshed [%s]", p_srv->description);
}
}
}
}
} else { /* LOCK */
p_srv->score.value = CLAMP(si->score.value, SCORE_DOWN, SCORE_MAX);
if (p_srv->locked) {
GRID_TRACE2("SRV already locked [%s]", p_srv->description);
} else {
p_srv->locked = TRUE;
GRID_TRACE2("SRV locked [%s]", p_srv->description);
}
}
return p_srv;
}
GError*
_open_and_lock(struct meta1_backend_s *m1, struct oio_url_s *url,
enum m1v2_open_type_e how, struct sqlx_sqlite3_s **handle)
{
EXTRA_ASSERT(m1 != NULL);
EXTRA_ASSERT(url != NULL);
EXTRA_ASSERT(handle != NULL);
GRID_TRACE2("%s(%p,%p,%d,%p)", __FUNCTION__, (void*)m1,
oio_url_get (url, OIOURL_HEXID), how, (void*)handle);
if (!oio_url_has (url, OIOURL_HEXID))
return NEWERROR (CODE_BAD_REQUEST, "Partial URL (missing HEXID)");
if (!m1b_check_ns_url (m1, url))
return NEWERROR(CODE_NAMESPACE_NOTMANAGED, "Invalid NS");
gchar base[5];
const guint8 *cid = oio_url_get_id(url);
g_snprintf(base, sizeof(base), "%02X%02X", cid[0], cid[1]);
if (!meta1_prefixes_is_managed(m1->prefixes, cid))
return NEWERROR(CODE_RANGE_NOTFOUND, "prefix [%s] not managed", base);
/* Now open/lock the base in a way suitable for our op */
struct sqlx_name_s n = {.base=base, .type=NAME_SRVTYPE_META1, .ns=m1->ns_name};
GError *err = sqlx_repository_open_and_lock(m1->repo, &n, m1_to_sqlx(how), handle, NULL);
if (err != NULL) {
if (!CODE_IS_REDIRECT(err->code))
g_prefix_error(&err, "Open/Lock error: ");
return err;
}
EXTRA_ASSERT(*handle != NULL);
GRID_TRACE("Opened and locked [%s][%s] -> [%s][%s]",
base, NAME_SRVTYPE_META1,
(*handle)->name.base, (*handle)->name.type);
return NULL;
}
GError*
__create_user(struct sqlx_sqlite3_s *sq3, struct oio_url_s *url)
{
if (!oio_url_has_fq_container (url))
return NEWERROR(CODE_BAD_REQUEST, "Partial URL");
static const gchar *sql = "INSERT INTO users ('cid','account','user') VALUES (?,?,?)";
GError *err = NULL;
sqlite3_stmt *stmt = NULL;
int rc;
EXTRA_ASSERT(sq3 != NULL);
EXTRA_ASSERT(sq3->db != NULL);
/* Prepare the statement */
sqlite3_prepare_debug(rc, sq3->db, sql, -1, &stmt, NULL);
if (rc != SQLITE_OK)
err = M1_SQLITE_GERROR(sq3->db, rc);
else {
sqlite3_bind_blob(stmt, 1, oio_url_get_id(url), oio_url_get_id_size(url), NULL);
sqlite3_bind_text(stmt, 2, oio_url_get(url, OIOURL_ACCOUNT), -1, NULL);
sqlite3_bind_text(stmt, 3, oio_url_get(url, OIOURL_USER), -1, NULL);
/* Run the results */
do { rc = sqlite3_step(stmt); } while (rc == SQLITE_ROW);
if (rc != SQLITE_OK && rc != SQLITE_DONE) {
err = M1_SQLITE_GERROR(sq3->db, rc);
if (rc == SQLITE_CONSTRAINT) {
g_prefix_error(&err, "Already created? ");
err->code = CODE_CONTAINER_EXISTS;
}
}
sqlite3_finalize_debug(rc, stmt);
}
if (err)
GRID_DEBUG("User creation failed : (%d) %s", err->code, err->message);
return err;
}
GError*
__info_user(struct sqlx_sqlite3_s *sq3, struct oio_url_s *url, gboolean ac,
struct oio_url_s ***result)
{
GError *err = NULL;
sqlite3_stmt *stmt = NULL;
GPtrArray *gpa;
int rc;
gboolean found;
EXTRA_ASSERT(sq3 != NULL);
EXTRA_ASSERT(sq3->db != NULL);
EXTRA_ASSERT(url != NULL);
retry:
/* Prepare the statement */
sqlite3_prepare_debug(rc, sq3->db, "SELECT account,user FROM users WHERE cid = ?", -1, &stmt, NULL);
if (rc != SQLITE_OK)
return M1_SQLITE_GERROR(sq3->db, rc);
(void) sqlite3_bind_blob(stmt, 1, oio_url_get_id (url), oio_url_get_id_size (url), NULL);
/* Run the results */
found = FALSE;
gpa = result ? g_ptr_array_new() : NULL;
do { if (SQLITE_ROW == (rc = sqlite3_step(stmt))) {
found = TRUE;
if (!gpa) continue;
struct oio_url_s *u = oio_url_empty ();
oio_url_set (u, OIOURL_NS, oio_url_get (url, OIOURL_NS));
oio_url_set (u, OIOURL_ACCOUNT, (char*)sqlite3_column_text(stmt, 0));
oio_url_set (u, OIOURL_USER, (char*)sqlite3_column_text(stmt, 1));
oio_url_set (u, OIOURL_HEXID, oio_url_get (url, OIOURL_HEXID));
g_ptr_array_add(gpa, u);
} } while (rc == SQLITE_ROW);
if (rc != SQLITE_DONE && rc != SQLITE_OK) {
err = M1_SQLITE_GERROR(sq3->db, rc);
g_prefix_error(&err, "DB error: ");
}
sqlite3_finalize_debug(rc,stmt);
stmt = NULL;
if (err) {
if (gpa) {
g_ptr_array_set_free_func (gpa, (GDestroyNotify)oio_url_clean);
g_ptr_array_free (gpa, TRUE);
}
return err;
}
if (!found) {
if (gpa) g_ptr_array_free (gpa, TRUE);
if (ac) {
ac = FALSE; /* do not retry */
err = __create_user (sq3, url);
if (!err) goto retry;
}
return NEWERROR(CODE_USER_NOTFOUND, "no such container");
}
if (gpa)
*result = (struct oio_url_s**) metautils_gpa_to_array(gpa, TRUE);
return NULL;
}
GError *
sqlx_cache_open_and_lock_base(sqlx_cache_t *cache, const hashstr_t *hname,
gint *result)
{
gint bd;
GError *err = NULL;
sqlx_base_t *base = NULL;
EXTRA_ASSERT(cache != NULL);
EXTRA_ASSERT(hname != NULL);
EXTRA_ASSERT(result != NULL);
gint64 start = oio_ext_monotonic_time();
gint64 deadline = DEFAULT_CACHE_OPEN_TIMEOUT;
if (cache->open_timeout > 0)
deadline = cache->open_timeout;
GRID_TRACE2("%s(%p,%s,%p) delay = %"G_GINT64_FORMAT, __FUNCTION__,
(void*)cache, hname ? hashstr_str(hname) : "NULL",
(void*)result, deadline);
deadline += start;
g_mutex_lock(&cache->lock);
cache->used = TRUE;
retry:
bd = sqlx_lookup_id(cache, hname);
if (bd < 0) {
if (!(err = sqlx_base_reserve(cache, hname, &base))) {
bd = base->index;
*result = base->index;
sqlx_base_debug("OPEN", base);
}
else {
GRID_DEBUG("No base available for [%s] (%d %s)",
hashstr_str(hname), err->code, err->message);
if (sqlx_expire_first_idle_base(cache, 0) >= 0) {
g_clear_error(&err);
goto retry;
}
}
}
else {
base = GET(cache, bd);
gint64 now = oio_ext_monotonic_time ();
if (now > deadline) {
err = NEWERROR (CODE_UNAVAILABLE,
"DB busy (after %"G_GINT64_FORMAT" ms)",
(now - start) / G_TIME_SPAN_MILLISECOND);
} else switch (base->status) {
case SQLX_BASE_FREE:
EXTRA_ASSERT(base->count_open == 0);
EXTRA_ASSERT(base->owner == NULL);
GRID_ERROR("free base referenced");
g_assert_not_reached();
break;
case SQLX_BASE_IDLE:
case SQLX_BASE_IDLE_HOT:
EXTRA_ASSERT(base->count_open == 0);
EXTRA_ASSERT(base->owner == NULL);
sqlx_base_move_to_list(cache, base, SQLX_BASE_USED);
base->count_open ++;
base->owner = g_thread_self();
*result = base->index;
break;
case SQLX_BASE_USED:
EXTRA_ASSERT(base->count_open > 0);
EXTRA_ASSERT(base->owner != NULL);
if (base->owner != g_thread_self()) {
GRID_DEBUG("Base [%s] in use by another thread (%X), waiting...",
hashstr_str(hname), oio_log_thread_id(base->owner));
/* The lock is held by another thread/request.
XXX(jfs): do not use 'now' because it can be a fake clock */
g_cond_wait_until(base->cond, &cache->lock,
g_get_monotonic_time() + oio_cache_period_cond_wait);
goto retry;
}
base->owner = g_thread_self();
base->count_open ++;
*result = base->index;
break;
case SQLX_BASE_CLOSING:
EXTRA_ASSERT(base->owner != NULL);
/* Just wait for a notification then retry
XXX(jfs): do not use 'now' because it can be a fake clock */
g_cond_wait_until(base->cond, &cache->lock,
g_get_monotonic_time() + oio_cache_period_cond_wait);
goto retry;
}
}
if (base) {
if (!err) {
sqlx_base_debug(__FUNCTION__, base);
EXTRA_ASSERT(base->owner == g_thread_self());
EXTRA_ASSERT(base->count_open > 0);
}
g_cond_signal(base->cond);
}
g_mutex_unlock(&cache->lock);
return err;
}
static enum http_rc_e
_registration (struct req_args_s *args, enum reg_op_e op, struct json_object *jsrv)
{
GError *err;
if (!jsrv || !json_object_is_type (jsrv, json_type_object))
return _reply_common_error (args, BADREQ("Expected: json object"));
if (!push_queue)
return _reply_bad_gateway(args, SYSERR("Service upstream disabled"));
if (NULL != (err = _cs_check_tokens(args)))
return _reply_notfound_error (args, err);
struct service_info_s *si = NULL;
err = service_info_load_json_object (jsrv, &si, TRUE);
if (err) {
g_prefix_error (&err, "JSON error: ");
if (err->code == CODE_BAD_REQUEST)
return _reply_format_error (args, err);
else
return _reply_system_error (args, err);
}
if (!si->type[0]) {
service_info_clean (si);
return _reply_format_error (args, BADREQ("Service type not specified"));
}
if (!si->ns_name[0]) {
GRID_TRACE2("%s NS forced to %s", __FUNCTION__, si->ns_name);
g_strlcpy (si->ns_name, nsname, sizeof(si->ns_name));
} else if (!validate_namespace (si->ns_name)) {
service_info_clean (si);
return _reply_format_error (args, BADNS());
}
gchar *k = service_info_key (si);
STRING_STACKIFY(k);
GRID_TRACE2("%s op=%s score=%d key=[%s]", __FUNCTION__,
_regop_2str(op), si->score.value, k);
switch (op) {
case REGOP_PUSH:
si->score.value = SCORE_UNSET;
if (!service_is_known (k)) {
service_learn (k);
service_tag_set_value_boolean (service_info_ensure_tag (
si->tags, NAME_TAGNAME_RAWX_FIRST), TRUE);
}
break;
case REGOP_LOCK:
si->score.value = CLAMP(si->score.value, SCORE_DOWN, SCORE_MAX);
break;
case REGOP_UNLOCK:
si->score.value = SCORE_UNLOCK;
break;
default:
g_assert_not_reached();
}
if (cs_expire_local_services > 0) {
struct service_info_s *v = service_info_dup (si);
v->score.timestamp = oio_ext_monotonic_seconds ();
PUSH_DO(
const struct service_info_s *si0 = lru_tree_get(srv_registered, k);
if (si0) v->score.value = si0->score.value;
lru_tree_insert (srv_registered, g_strdup(k), v);
);
}
static GError*
_cache_load_from_m0(struct meta1_prefixes_set_s *m1ps,
const gchar *ns_name,
const struct addr_info_s *local_addr,
struct addr_info_s *m0_addr,
GArray **updated_prefixes,
gboolean *meta0_ok)
{
GError *err = NULL;
GSList *m0info_list = NULL;
EXTRA_ASSERT(m1ps != NULL);
GRID_TRACE2("%s(%p,%s,%p,%p)", __FUNCTION__, m1ps, ns_name, local_addr,
m0_addr);
(void)ns_name;
gchar m0[STRLEN_ADDRINFO];
grid_addrinfo_to_string (m0_addr, m0, sizeof(m0));
err = meta0_remote_get_meta1_all(m0, &m0info_list);
if (err) {
g_prefix_error(&err, "Remote error: ");
return err;
}
if (!m0info_list) {
GRID_DEBUG("META0 has no prefix configured!");
return NULL;
}
*meta0_ok = TRUE;
guint8 *cache = _cache_from_m0l(m0info_list, local_addr);
GPtrArray *by_prefix = meta0_utils_list_to_array(m0info_list);
g_mutex_lock(&m1ps->lock);
GRID_DEBUG("Got %u prefixes from M0, %u in place",
by_prefix->len, m1ps->by_prefix ? m1ps->by_prefix->len : 0);
if ( m1ps->by_prefix ) {
guint prefix;
*updated_prefixes = g_array_new(FALSE, FALSE, sizeof(guint16));
for( prefix=0 ; prefix <65536 ;prefix++) {
if ( _cache_is_managed(m1ps->cache,(guint8 *)&prefix) != _cache_is_managed( cache,(guint8 *)&prefix)) {
g_array_append_vals(*updated_prefixes, &prefix, 1);
}
}
}
SWAP_PTR(m1ps->by_prefix, by_prefix);
SWAP_PTR(m1ps->cache, cache);
g_mutex_unlock(&m1ps->lock);
if (by_prefix)
meta0_utils_array_clean(by_prefix);
by_prefix = NULL;
if (cache)
g_free(cache);
cache = NULL;
g_slist_foreach(m0info_list, meta0_info_gclean, NULL);
g_slist_free(m0info_list);
return NULL;
}
GError *
sqlx_cache_open_and_lock_base(sqlx_cache_t *cache, const hashstr_t *hname,
gint *result)
{
gint bd;
GError *err = NULL;
sqlx_base_t *base = NULL;
GRID_TRACE2("%s(%p,%s,%p)", __FUNCTION__, (void*)cache,
hname ? hashstr_str(hname) : "NULL", (void*)result);
EXTRA_ASSERT(cache != NULL);
EXTRA_ASSERT(hname != NULL);
EXTRA_ASSERT(result != NULL);
gint64 deadline = g_get_monotonic_time();
if (cache->open_timeout >= 0) {
deadline += cache->open_timeout * G_TIME_SPAN_MILLISECOND;
} else {
deadline += 5 * G_TIME_SPAN_MINUTE;
}
g_mutex_lock(&cache->lock);
cache->used = TRUE;
retry:
bd = sqlx_lookup_id(cache, hname);
if (bd < 0) {
if (!(err = sqlx_base_reserve(cache, hname, &base))) {
bd = base->index;
*result = base->index;
sqlx_base_debug("OPEN", base);
}
else {
GRID_DEBUG("No base available for [%s] (%d %s)",
hashstr_str(hname), err->code, err->message);
if (sqlx_expire_first_idle_base(cache, NULL) >= 0) {
g_clear_error(&err);
goto retry;
}
}
}
else {
base = GET(cache, bd);
switch (base->status) {
case SQLX_BASE_FREE:
EXTRA_ASSERT(base->count_open == 0);
EXTRA_ASSERT(base->owner == NULL);
GRID_ERROR("free base referenced");
g_assert_not_reached();
break;
case SQLX_BASE_IDLE:
case SQLX_BASE_IDLE_HOT:
EXTRA_ASSERT(base->count_open == 0);
EXTRA_ASSERT(base->owner == NULL);
sqlx_base_move_to_list(cache, base, SQLX_BASE_USED);
base->count_open ++;
base->owner = g_thread_self();
*result = base->index;
break;
case SQLX_BASE_USED:
EXTRA_ASSERT(base->count_open > 0);
EXTRA_ASSERT(base->owner != NULL);
if (base->owner != g_thread_self()) {
GRID_DEBUG("Base [%s] in use by another thread (%X), waiting...",
hashstr_str(hname), oio_log_thread_id(base->owner));
// The lock is held by another thread/request
if (g_cond_wait_until(base->cond, &cache->lock, deadline)) {
GRID_DEBUG("Retrying to open [%s]", hashstr_str(hname));
goto retry;
} else {
if (cache->open_timeout > 0) {
err = NEWERROR(CODE_UNAVAILABLE,
"database currently in use by another request"
" (we waited %ldms)",
cache->open_timeout);
} else {
err = NEWERROR(CODE_UNAVAILABLE,
"database currently in use by another request");
}
GRID_DEBUG("failed to open base: "
"in use by another request (thread %X)",
oio_log_thread_id(base->owner));
break;
}
}
base->owner = g_thread_self();
base->count_open ++;
*result = base->index;
break;
case SQLX_BASE_CLOSING:
EXTRA_ASSERT(base->owner != NULL);
// Just wait for a notification then retry
if (g_cond_wait_until(base->cond, &cache->lock, deadline))
goto retry;
else {
err = NEWERROR(CODE_UNAVAILABLE,
"Database stuck in closing state");
break;
}
}
}
if (base) {
if (!err) {
sqlx_base_debug(__FUNCTION__, base);
EXTRA_ASSERT(base->owner == g_thread_self());
EXTRA_ASSERT(base->count_open > 0);
}
g_cond_signal(base->cond);
}
g_mutex_unlock(&cache->lock);
return err;
}