jmoses-couchbase 1.3.6

data/ext/couchbase_ext/multithread_plugin.c

@@ -0,0 +1,1090 @@
+/* vim: ft=c et ts=8 sts=4 sw=4 cino=
+ *
+ *   Copyright 2012 Couchbase, Inc.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include "couchbase_ext.h"
+
+
+#ifndef _WIN32
+
+#ifndef HAVE_RB_THREAD_BLOCKING_REGION
+#include <rubysig.h>
+#endif
+#include <errno.h>
+#ifdef HAVE_POLL
+#include <poll.h>
+#endif
+
+/* events sorted array */
+typedef struct rb_mt_event rb_mt_event;
+struct rb_mt_event {
+    void *cb_data;
+    void (*handler)(lcb_socket_t sock, short which, void *cb_data);
+    lcb_socket_t socket;
+    int loop_index;
+    short flags;
+    short actual_flags;
+    short inserted;
+    rb_mt_event *next;
+};
+
+typedef struct rb_mt_socket_list rb_mt_socket_list;
+struct rb_mt_socket_list {
+    lcb_socket_t socket;
+    short flags;
+    rb_mt_event *first;
+};
+
+typedef struct rb_mt_events rb_mt_events;
+struct rb_mt_events {
+    uint32_t capa;
+    uint32_t count;
+    rb_mt_socket_list *sockets;
+};
+
+    static int
+events_init(rb_mt_events *events)
+{
+    rb_mt_socket_list *new_socks = malloc(4 * sizeof(*new_socks));
+    if (new_socks == NULL) {
+        return 0;
+    }
+    events->capa = 4;
+    events->count = 0;
+    events->sockets = new_socks;
+    return 1;
+}
+
+    static void
+events_finalize(rb_mt_events *events)
+{
+    if (events->sockets) {
+        uint32_t i;
+        for(i = 0; i < events->count; i++) {
+            rb_mt_socket_list *list = &events->sockets[i];
+            while(list->first) {
+                rb_mt_event *next = list->first->next;
+                free(list->first);
+                list->first = next;
+            }
+        }
+        free(events->sockets);
+        events->sockets = NULL;
+    }
+    events->capa = 0;
+    events->count = 0;
+}
+
+    static uint32_t
+events_index(rb_mt_events *events, lcb_socket_t socket)
+{
+    uint32_t m, l = 0, r = events->count;
+    while(l < r) {
+        m = l + (r - l) / 2;
+        if (events->sockets[m].socket >= socket) {
+            r = m;
+        } else {
+            l = m + 1;
+        }
+    }
+    return l;
+}
+
+    static void
+events_insert(rb_mt_events *events, rb_mt_event *event)
+{
+    uint32_t i = events_index(events, event->socket);
+    rb_mt_socket_list *list = &events->sockets[i];
+    if (i == events->count || list->socket != event->socket) {
+        if (events->capa == events->count) {
+            uint32_t new_capa = events->capa << 1;
+            rb_mt_socket_list *new_socks = realloc(events->sockets, new_capa * sizeof(*new_socks));
+            if (new_socks == NULL) {
+                rb_raise(cb_eClientNoMemoryError, "failed to allocate memory for events array");
+            }
+            events->sockets = new_socks;
+            events->capa = new_capa;
+            list = &events->sockets[i];
+        }
+        if (i < events->count) {
+            MEMMOVE(events->sockets+i+1, events->sockets+i, rb_mt_socket_list, events->count - i);
+        }
+        events->count++;
+        list->socket = event->socket;
+        list->flags = event->flags;
+        list->first = event;
+        event->next = NULL;
+    } else {
+        list->flags |= event->flags;
+        event->next = list->first;
+        list->first = event;
+    }
+    event->inserted = 1;
+}
+
+    static void
+event_list_fix_flags(rb_mt_socket_list *list)
+{
+    short flags = 0;
+    rb_mt_event *event = list->first;
+    while (event) {
+        flags |= event->flags;
+        event = event->next;
+    }
+    list->flags = flags;
+}
+
+    static void
+events_remove(rb_mt_events *events, rb_mt_event *event)
+{
+    uint32_t i = events_index(events, event->socket);
+    rb_mt_socket_list *list = &events->sockets[i];
+    rb_mt_event **next;
+    if (list->socket != event->socket) {
+        rb_raise(rb_eIndexError, "There is no socket in event loop");
+    }
+    next = &list->first;
+    for(;;) {
+        if (*next == NULL) {
+            rb_raise(rb_eIndexError, "There is no event in event loop");
+        }
+        if (*next == event) {
+            *next = event->next;
+            event->next = NULL;
+            event->inserted = 0;
+            break;
+        }
+        next = &event->next;
+    }
+    if (list->first == NULL) {
+        MEMMOVE(events->sockets + i, events->sockets + i + 1, rb_mt_socket_list, events->count - i - 1);
+        events->count--;
+    } else {
+        event_list_fix_flags(list);
+    }
+}
+
+    static void
+events_fix_flags(rb_mt_events *events, lcb_socket_t socket)
+{
+    uint32_t i = events_index(events, socket);
+    rb_mt_socket_list *list = &events->sockets[i];
+    if (list->socket != socket) {
+        rb_raise(rb_eIndexError, "There is no socket in event loop");
+    }
+    event_list_fix_flags(list);
+}
+
+    static inline lcb_socket_t
+events_max_fd(rb_mt_events *events)
+{
+    if (events->count) {
+        return events->sockets[events->count - 1].socket;
+    } else {
+        return -1;
+    }
+}
+
+/* events sorted array end */
+
+/* timers heap */
+typedef struct rb_mt_timer rb_mt_timer;
+struct rb_mt_timer {
+    void *cb_data;
+    void (*handler)(lcb_socket_t sock, short which, void *cb_data);
+    int index;
+    hrtime_t ts;
+    hrtime_t period;
+};
+
+typedef struct rb_mt_timers rb_mt_timers;
+struct rb_mt_timers {
+    uint32_t capa;
+    uint32_t count;
+    rb_mt_timer **timers;
+};
+
+    static int
+timers_init(rb_mt_timers *timers)
+{
+    rb_mt_timer **new_timers = malloc(4 * sizeof(*new_timers));
+    if (new_timers == NULL) {
+        return 0;
+    }
+    timers->capa = 4;
+    timers->count = 0;
+    timers->timers = new_timers;
+    return 1;
+}
+
+    static void
+timers_finalize(rb_mt_timers *timers)
+{
+    if (timers->timers) {
+        uint32_t i;
+        for(i = 0; i < timers->count; i++) {
+            free(timers->timers[i]);
+        }
+        free(timers->timers);
+        timers->timers = NULL;
+    }
+    timers->count = 0;
+    timers->capa = 0;
+}
+
+#define tms_at(_timers, at) (_timers)->timers[(at)]
+#define tms_ts_at(timers, at) tms_at((timers), (at))->ts
+
+    static void
+timers_move_last(rb_mt_timers *timers, uint32_t to)
+{
+    if (to < timers->count - 1) {
+        rb_mt_timer *last = tms_at(timers, timers->count - 1);
+        tms_at(timers, to) = last;
+        last->index = to;
+    }
+    timers->count--;
+}
+
+    static inline void
+timers_swap(rb_mt_timers *timers, uint32_t i, uint32_t j)
+{
+    rb_mt_timer *itmp = tms_at(timers, j);
+    rb_mt_timer *jtmp = tms_at(timers, i);
+    tms_at(timers, i) = itmp;
+    tms_at(timers, j) = jtmp;
+    itmp->index = i;
+    jtmp->index = j;
+}
+
+static void timers_heapify_up(rb_mt_timers *timers, uint32_t pos);
+
+    static void
+timers_insert(rb_mt_timers *timers, rb_mt_timer *timer)
+{
+    if (timers->count == timers->capa) {
+        rb_mt_timer **new_timers;
+        size_t new_capa = timers->capa << 1;
+        new_timers = realloc(timers->timers, new_capa * sizeof(rb_mt_timer*));
+        if (new_timers == NULL) {
+            rb_raise(cb_eClientNoMemoryError, "failed to allocate memory for timers heap");
+        }
+        timers->timers = new_timers;
+        timers->capa = new_capa;
+    }
+    tms_at(timers, timers->count) = timer;
+    timer->index = timers->count;
+    timers->count++;
+    timers_heapify_up(timers, timer->index);
+}
+
+    static void
+timers_heapify_up(rb_mt_timers *timers, uint32_t pos)
+{
+    hrtime_t cur_ts = tms_ts_at(timers, pos);
+    uint32_t higher = (pos - 1) / 2;
+    while (pos && tms_ts_at(timers, higher) > cur_ts) {
+        timers_swap(timers, higher, pos);
+        pos = higher;
+        higher = (pos - 1) / 2;
+    }
+}
+
+    static void
+timers_heapify_down(rb_mt_timers *timers, uint32_t pos)
+{
+    uint32_t count = timers->count;
+    uint32_t middle = (timers->count - 2) / 2;
+    hrtime_t cur_ts = tms_ts_at(timers, pos);
+    if (count == 1) return;
+    while (pos <= middle) {
+        uint32_t min_pos = pos;
+        hrtime_t ch_ts, min_ts = cur_ts;
+
+        if ((ch_ts = tms_ts_at(timers, pos * 2 + 1)) < min_ts) {
+            min_pos = pos * 2 + 1;
+            min_ts = ch_ts;
+        }
+
+        if (pos * 2 + 2 < count && tms_ts_at(timers, pos * 2 + 2) < min_ts) {
+            min_pos = pos * 2 + 2;
+        }
+
+        if (min_pos == pos) break;
+        timers_swap(timers, pos, min_pos);
+        pos = min_pos;
+    }
+}
+
+    static void
+timers_heapify_item(rb_mt_timers *timers, uint32_t pos)
+{
+    if (pos && tms_ts_at(timers, pos) < tms_ts_at(timers, (pos - 1) / 2)) {
+        timers_heapify_up(timers, pos);
+    } else {
+        timers_heapify_down(timers, pos);
+    }
+}
+
+    static inline hrtime_t
+timers_minimum(rb_mt_timers *timers)
+{
+    if (timers->count) {
+        return tms_ts_at(timers, 0);
+    } else {
+        return 0;
+    }
+}
+
+    static inline rb_mt_timer *
+timers_first(rb_mt_timers *timers)
+{
+    if (timers->count) {
+        return tms_at(timers, 0);
+    } else {
+        return 0;
+    }
+}
+
+    static void
+timers_remove_timer(rb_mt_timers *timers, rb_mt_timer *timer)
+{
+    uint32_t at = timer->index;
+    timer->index = -1;
+    if (at < timers->count - 1) {
+        timers_move_last(timers, at);
+        timers_heapify_item(timers, at);
+    } else {
+        timers->count--;
+    }
+}
+
+    static void
+timers_run(rb_mt_timers *timers, hrtime_t now)
+{
+    hrtime_t next_time = timers_minimum(timers);
+    while (next_time && next_time < now) {
+        rb_mt_timer *first = timers_first(timers);
+
+        first->ts = now + first->period;
+        timers_heapify_item(timers, 0);
+
+        first->handler(-1, 0, first->cb_data);
+
+        next_time = timers_minimum(timers);
+    }
+}
+/* timers heap end */
+
+/* callbacks array */
+typedef struct rb_mt_callbacks rb_mt_callbacks;
+struct rb_mt_callbacks {
+    uint32_t capa;
+    uint32_t count;
+    rb_mt_event **events;
+};
+
+    static int
+callbacks_init(rb_mt_callbacks *callbacks)
+{
+    rb_mt_event **new_events = calloc(4, sizeof(*new_events));
+    if (new_events == NULL) {
+        return 0;
+    }
+    callbacks->events = new_events;
+    callbacks->capa = 4;
+    callbacks->count = 0;
+    return 1;
+}
+
+    static void
+callbacks_finalize(rb_mt_callbacks *callbacks)
+{
+    if (callbacks->events) {
+        free(callbacks->events);
+        callbacks->events = NULL;
+    }
+    callbacks->capa = 0;
+    callbacks->count = 0;
+}
+
+    static void
+callbacks_push(rb_mt_callbacks *callbacks, rb_mt_event *event)
+{
+    if (callbacks->count == callbacks->capa) {
+        uint32_t new_capa = callbacks->capa << 1;
+        rb_mt_event **new_events = realloc(callbacks->events, new_capa * sizeof(*new_events));
+        if (new_events == NULL) {
+            rb_raise(cb_eClientNoMemoryError, "failed to allocate memory for callbacks array");
+        }
+        callbacks->capa = new_capa;
+        callbacks->events = new_events;
+    }
+    event->loop_index = callbacks->count;
+    callbacks->events[callbacks->count] = event;
+    callbacks->count++;
+}
+
+    static void
+callbacks_remove(rb_mt_callbacks *callbacks, rb_mt_event *event)
+{
+    int i = event->loop_index;
+    if (i >= 0) {
+        if (callbacks->events[i] != event) {
+            rb_raise(rb_eIndexError, "callback index belongs to different callback");
+        }
+        event->loop_index = -1;
+        callbacks->events[i] = NULL;
+    }
+}
+
+    static void
+callbacks_run(rb_mt_callbacks *callbacks)
+{
+    uint32_t i;
+    for(i = 0; i < callbacks->count; i++) {
+        rb_mt_event *cb = callbacks->events[i];
+        if (cb) {
+            cb->loop_index = -1;
+            callbacks->events[i] = NULL;
+            cb->handler(cb->socket, cb->actual_flags, cb->cb_data);
+        }
+    }
+    callbacks->count = 0;
+}
+
+    static void
+callbacks_clean(rb_mt_callbacks *callbacks)
+{
+    uint32_t i;
+    for(i = 0; i < callbacks->count; i++) {
+        if (callbacks->events[i]) {
+            callbacks->events[i]->loop_index = -1;
+            callbacks->events[i] = NULL;
+        }
+    }
+    callbacks->count = 0;
+}
+/* callbacks array end */
+
+typedef struct rb_mt_loop rb_mt_loop;
+struct rb_mt_loop {
+    rb_mt_events events;
+    rb_mt_timers timers;
+    rb_mt_callbacks callbacks;
+    short run;
+};
+
+    static rb_mt_loop*
+loop_create()
+{
+    rb_mt_loop *loop = calloc(1, sizeof(*loop));
+    if (loop == NULL) return NULL;
+    if (!events_init(&loop->events)) goto free_loop;
+    if (!timers_init(&loop->timers)) goto free_events;
+    if (!callbacks_init(&loop->callbacks)) goto free_timers;
+    return loop;
+
+free_timers:
+    timers_finalize(&loop->timers);
+free_events:
+    events_finalize(&loop->events);
+free_loop:
+    free(loop);
+    return NULL;
+}
+
+    static void
+loop_destroy(rb_mt_loop *loop)
+{
+    events_finalize(&loop->events);
+    timers_finalize(&loop->timers);
+    callbacks_finalize(&loop->callbacks);
+    free(loop);
+}
+
+    static void
+loop_remove_event(rb_mt_loop *loop, rb_mt_event *event)
+{
+    if (event->inserted) {
+        events_remove(&loop->events, event);
+    }
+    callbacks_remove(&loop->callbacks, event);
+}
+
+    static void
+loop_enque_events(rb_mt_callbacks *callbacks, rb_mt_event *sock, short flags)
+{
+    while (sock) {
+        short actual = sock->flags & flags;
+        if (actual) {
+            sock->actual_flags = actual;
+            callbacks_push(callbacks, (rb_mt_event*)sock);
+        }
+        sock = sock->next;
+    }
+}
+
+/* loop select implementation */
+#ifndef HAVE_RB_THREAD_FD_SELECT
+typedef fd_set rb_fdset_t;
+#define rb_fd_init   FD_ZERO
+#define rb_fd_set    FD_SET
+#define rb_fd_isset  FD_ISSET
+#define rb_fd_term(set)  (void)0
+#define rb_thread_fd_select rb_thread_select
+#endif
+
+typedef struct loop_select_arg {
+    rb_mt_loop *loop;
+    rb_fdset_t in, out, exc;
+} ls_arg;
+
+    static void
+ls_arg_free(void *p) {
+    ls_arg *args = p;
+    if (args) {
+        rb_fd_term(&args->in);
+        rb_fd_term(&args->out);
+        xfree(args);
+    }
+}
+
+    static VALUE
+ls_arg_alloc(ls_arg **args)
+{
+    return Data_Make_Struct(rb_cObject, ls_arg, 0, ls_arg_free, *args);
+}
+
+    static VALUE
+loop_run_select(VALUE argp)
+{
+    ls_arg *args = (ls_arg*) argp;
+    rb_mt_loop *loop = args->loop;
+    rb_fdset_t *in = NULL, *out = NULL, *exc = NULL;
+    struct timeval timeout;
+    struct timeval *timeoutp = NULL;
+    int result, max = 0;
+    hrtime_t now, next_time;
+
+    next_time = timers_minimum(&loop->timers);
+    if (next_time) {
+        now = gethrtime();
+        if (next_time <= now) {
+            timeout.tv_sec = 0;
+            timeout.tv_usec = 0;
+        } else {
+            hrtime_t hrto = (next_time - now) / 1000;
+            timeout.tv_sec = (long)(hrto / 1000000);
+            timeout.tv_usec = (long)(hrto % 1000000);
+        }
+        timeoutp = &timeout;
+    }
+
+    if (loop->events.count) {
+        uint32_t i;
+        rb_fd_init(&args->in);
+        rb_fd_init(&args->out);
+        rb_fd_init(&args->exc);
+        for(i = 0; i < loop->events.count; i++) {
+            rb_mt_socket_list *list = &loop->events.sockets[i];
+            if (list->flags != 0) {
+                if (list->flags & LCB_READ_EVENT) {
+                    in = &args->in;
+                    rb_fd_set(list->socket, in);
+                }
+                if (list->flags & LCB_WRITE_EVENT) {
+                    out = &args->out;
+                    rb_fd_set(list->socket, out);
+                }
+                exc = &args->exc;
+                rb_fd_set(list->socket, exc);
+            }
+        }
+        max = events_max_fd(&loop->events) + 1;
+    }
+
+    result = rb_thread_fd_select(max, in, out, exc, timeoutp);
+
+    if (result < 0) {
+        rb_sys_fail("rb_thread_fd_select");
+    }
+    /* fix current time so that socket callbacks will not cause timers timeouts */
+    if (next_time) {
+        now = gethrtime();
+    }
+
+    if (result > 0) {
+        uint32_t i;
+        for(i = 0; i < loop->events.count && result; i++) {
+            rb_mt_socket_list *list = loop->events.sockets + i;
+            rb_mt_event *sock = list->first;
+            short flags = 0;
+            if (in && rb_fd_isset(list->socket, in)) {
+                flags |= LCB_READ_EVENT;
+                result--;
+            }
+            if (out && rb_fd_isset(list->socket, out)) {
+                flags |= LCB_WRITE_EVENT;
+                result--;
+            }
+            if (exc && rb_fd_isset(list->socket, exc)) {
+                flags = LCB_ERROR_EVENT | LCB_WRITE_EVENT;
+                result--;
+            }
+            if (flags) {
+                loop_enque_events(&loop->callbacks, sock, flags);
+            }
+        }
+        callbacks_run(&loop->callbacks);
+    }
+
+    if (next_time) {
+        timers_run(&loop->timers, now);
+    }
+    if (loop->events.count == 0 && loop->timers.count == 0) {
+        loop->run = 0;
+    }
+    return Qnil;
+}
+
+    static VALUE
+loop_select_cleanup(VALUE argp)
+{
+    ls_arg *args = DATA_PTR(argp);
+    if (args) {
+        callbacks_clean(&args->loop->callbacks);
+        ls_arg_free(args);
+        DATA_PTR(argp) = 0;
+    }
+    return Qnil;
+}
+/* loop select implementaion end */
+
+/* loop poll implementation */
+#ifdef HAVE_POLL
+/* code influenced by ruby's source and cool.io */
+#define POLLIN_SET (POLLIN | POLLHUP | POLLERR)
+#define POLLOUT_SET (POLLOUT | POLLHUP | POLLERR)
+#define HRTIME_INFINITY ((hrtime_t)~0)
+
+#ifdef HAVE_PPOLL
+    static int
+xpoll(struct pollfd *fds, nfds_t nfds, hrtime_t timeout)
+{
+    if (timeout != HRTIME_INFINITY) {
+        struct timespec ts;
+        ts.tv_sec = (long)(timeout / (1000 * 1000 * 1000));
+        ts.tv_nsec = (long)(timeout % (1000 * 1000 * 1000));
+        return ppoll(fds, nfds, &ts, NULL);
+    }
+    return ppoll(fds, nfds, NULL, NULL);
+}
+#else
+#define TIMEOUT_MAX ((hrtime_t)(((unsigned int)~0) >> 1))
+    static int
+xpoll(struct pollfd *fds, nfds_t nfds, hrtime_t timeout)
+{
+    int ts = -1;
+    if (timeout != HRTIME_INFINITY) {
+        timeout = (timeout + 999999) / (1000 * 1000);
+        if (timeout <= TIMEOUT_MAX) {
+            ts = (int)timeout;
+        }
+    }
+    return poll(fds, nfds, ts);
+}
+#endif
+
+typedef struct poll_args lp_arg;
+struct poll_args {
+    rb_mt_loop *loop;
+    struct pollfd *fds;
+    nfds_t nfd;
+    hrtime_t ts;
+    int result;
+    int lerrno;
+};
+
+    static void
+lp_arg_free(void *p)
+{
+    lp_arg *args = p;
+    if (args) {
+        if (args->fds) {
+            free(args->fds);
+        }
+        xfree(args);
+    }
+}
+
+    static VALUE
+lp_arg_alloc(lp_arg **args)
+{
+    return Data_Make_Struct(rb_cObject, lp_arg, 0, lp_arg_free, *args);
+}
+
+#if defined(HAVE_RB_THREAD_BLOCKING_REGION) || defined(HAVE_RB_THREAD_CALL_WITHOUT_GVL)
+    static VALUE
+loop_blocking_poll(void *argp)
+{
+    lp_arg *args = argp;
+    args->result = xpoll(args->fds, args->nfd, args->ts);
+    if (args->result < 0) args->lerrno = errno;
+    return Qnil;
+}
+#endif
+
+    static VALUE
+loop_run_poll(VALUE argp)
+{
+    lp_arg *args = (lp_arg*)argp;
+    rb_mt_loop *loop = args->loop;
+    hrtime_t now, next_time;
+
+    if (loop->events.count) {
+        uint32_t i;
+        args->fds = calloc(loop->events.count, sizeof(struct pollfd));
+        if (args->fds == NULL) {
+            rb_raise(cb_eClientNoMemoryError, "failed to allocate memory for pollfd");
+        }
+        for(i = 0; i < loop->events.count; i++) {
+            rb_mt_socket_list *list = &loop->events.sockets[i];
+            args->fds[i].fd = list->socket;
+            args->fds[i].events =
+                (list->flags & LCB_READ_EVENT ? POLLIN : 0) |
+                (list->flags & LCB_WRITE_EVENT ? POLLOUT : 0);
+        }
+        args->nfd = loop->events.count;
+    }
+
+retry:
+    next_time = timers_minimum(&loop->timers);
+    if (next_time) {
+        now = gethrtime();
+        args->ts = next_time <= now ? 0 : next_time - now;
+    } else {
+        args->ts = HRTIME_INFINITY;
+    }
+
+
+#if defined(HAVE_RB_THREAD_CALL_WITHOUT_GVL)
+    rb_thread_call_without_gvl((void *(*)(void*))loop_blocking_poll, args, RUBY_UBF_IO, 0);
+#elif defined(HAVE_RB_THREAD_BLOCKING_REGION)
+    rb_thread_blocking_region(loop_blocking_poll, args, RUBY_UBF_PROCESS, NULL);
+#else
+    if (rb_thread_alone()) {
+        TRAP_BEG;
+        args->result = xpoll(args->fds, args->nfd, args->ts);
+        if (args->result < 0) args->lerrno = errno;
+        TRAP_END;
+    } else {
+        /* 5 millisecond pause */
+        hrtime_t mini_pause = 5000000;
+        int exact = 0;
+        if (args->ts != HRTIME_INFINITY && args->ts < mini_pause) {
+            mini_pause = args->ts;
+            exact = 1;
+        }
+        TRAP_BEG;
+        args->result = xpoll(args->fds, args->nfd, mini_pause);
+        if (args->result < 0) args->lerrno = errno;
+        TRAP_END;
+        if (args->result == 0 && !exact) {
+            args->result = -1;
+            args->lerrno = EINTR;
+        }
+    }
+#endif
+
+    if (args->result < 0) {
+        errno = args->lerrno;
+        switch (errno) {
+            case EINTR:
+#ifdef ERESTART
+            case ERESTART:
+#endif
+#ifndef HAVE_RB_THREAD_BLOCKING_REGION
+                rb_thread_schedule();
+#endif
+                goto retry;
+        }
+        rb_sys_fail("poll");
+        return Qnil;
+    }
+
+    if (next_time) {
+        now = gethrtime();
+    }
+
+    if (args->result > 0) {
+        uint32_t cnt = args->result;
+        uint32_t fd_n = 0, ev_n = 0;
+        while (cnt && fd_n < args->nfd && ev_n < loop->events.count) {
+            struct pollfd *res = args->fds + fd_n;
+            rb_mt_socket_list *list = loop->events.sockets + ev_n;
+            rb_mt_event *sock = list->first;
+
+            /* if plugin used correctly, this checks are noop */
+            if (res->fd < list->socket) {
+                fd_n++;
+                continue;
+            } else if (res->fd > list->socket) {
+                ev_n++;
+                continue;
+            }
+
+            if (res->revents) {
+                short flags =
+                    ((res->revents & POLLIN_SET) ? LCB_READ_EVENT : 0) |
+                    ((res->revents & POLLOUT_SET) ? LCB_WRITE_EVENT : 0);
+                cnt--;
+                loop_enque_events(&loop->callbacks, sock, flags);
+            }
+            fd_n++;
+            ev_n++;
+        }
+        callbacks_run(&loop->callbacks);
+    }
+
+    if (next_time) {
+        timers_run(&loop->timers, now);
+    }
+    if (loop->events.count == 0 && loop->timers.count == 0) {
+        loop->run = 0;
+    }
+    return Qnil;
+}
+
+    static VALUE
+loop_poll_cleanup(VALUE argp)
+{
+    lp_arg *args = DATA_PTR(argp);
+    if (args) {
+        callbacks_clean(&args->loop->callbacks);
+        lp_arg_free(args);
+        DATA_PTR(argp) = 0;
+    }
+    return Qnil;
+}
+#endif
+/* loop poll implementation end */
+
+    static void
+loop_run(rb_mt_loop *loop)
+{
+
+    loop->run = 1;
+
+    while(loop->run) {
+#ifdef HAVE_POLL
+        /* prefer use of poll when it gives some benefits, but use rb_thread_fd_select when it is sufficient */
+        lcb_socket_t max = events_max_fd(&loop->events);
+        int use_poll = max >= 128;
+        if (use_poll) {
+            lp_arg *args;
+            VALUE argp = lp_arg_alloc(&args);
+            args->loop = loop;
+            rb_ensure(loop_run_poll, (VALUE)args, loop_poll_cleanup, argp);
+        } else
+#endif
+        {
+            ls_arg *args;
+            VALUE argp = ls_arg_alloc(&args);
+            args->loop = loop;
+            rb_ensure(loop_run_select, (VALUE)args, loop_select_cleanup, argp);
+        }
+    }
+}
+
+    static void *
+lcb_io_create_event(struct lcb_io_opt_st *iops)
+{
+    rb_mt_event *event = calloc(1, sizeof(*event));
+    (void)iops;
+    event->loop_index = -1;
+    return event;
+}
+
+    static int
+lcb_io_update_event(struct lcb_io_opt_st *iops,
+        lcb_socket_t sock,
+        void *eventp,
+        short flags,
+        void *cb_data,
+        void (*handler)(lcb_socket_t sock,
+            short which,
+            void *cb_data))
+{
+    rb_mt_loop *loop = iops->v.v0.cookie;
+    rb_mt_event *event = eventp;
+    short old_flags = event->flags;
+
+    if (event->inserted && old_flags == flags &&
+            cb_data == event->cb_data && handler == event->handler)
+    {
+        return 0;
+    }
+    loop_remove_event(loop, event);
+    event->flags = flags;
+    event->cb_data = cb_data;
+    event->handler = handler;
+    event->socket = sock;
+    if (!event->inserted) {
+        events_insert(&loop->events, event);
+    }
+    if ((old_flags & flags) != old_flags) {
+        events_fix_flags(&loop->events, sock);
+    }
+    return 0;
+}
+
+    static void
+lcb_io_delete_event(struct lcb_io_opt_st *iops,
+        lcb_socket_t sock,
+        void *event)
+{
+    loop_remove_event((rb_mt_loop*)iops->v.v0.cookie, (rb_mt_event*)event);
+    (void)sock;
+}
+
+    static void
+lcb_io_destroy_event(struct lcb_io_opt_st *iops,
+        void *event)
+{
+    lcb_io_delete_event(iops, -1, event);
+    free(event);
+}
+
+    static void *
+lcb_io_create_timer(struct lcb_io_opt_st *iops)
+{
+    rb_mt_timer *timer = calloc(1, sizeof(*timer));
+    timer->index = -1;
+    (void)iops;
+    return timer;
+}
+
+    static int
+lcb_io_update_timer(struct lcb_io_opt_st *iops, void *event,
+        lcb_uint32_t usec, void *cb_data,
+        void (*handler)(lcb_socket_t sock, short which, void *cb_data))
+{
+    rb_mt_loop *loop = iops->v.v0.cookie;
+    rb_mt_timer *timer = event;
+
+    timer->period = usec * (hrtime_t)1000;
+    timer->ts = gethrtime() + timer->period;
+    timer->cb_data = cb_data;
+    timer->handler = handler;
+    if (timer->index != -1) {
+        timers_heapify_item(&loop->timers, timer->index);
+    } else {
+        timers_insert(&loop->timers, timer);
+    }
+    return 0;
+}
+
+    static void
+lcb_io_delete_timer(struct lcb_io_opt_st *iops, void *event)
+{
+    rb_mt_loop *loop = iops->v.v0.cookie;
+    rb_mt_timer *timer = event;
+    if (timer->index != -1) {
+        timers_remove_timer(&loop->timers, timer);
+    }
+}
+
+    static void
+lcb_io_destroy_timer(struct lcb_io_opt_st *iops, void *timer)
+{
+    lcb_io_delete_timer(iops, timer);
+    free(timer);
+}
+
+    static void
+lcb_io_stop_event_loop(struct lcb_io_opt_st *iops)
+{
+    rb_mt_loop *loop = iops->v.v0.cookie;
+    loop->run = 0;
+}
+
+    static void
+lcb_io_run_event_loop(struct lcb_io_opt_st *iops)
+{
+    rb_mt_loop *loop = iops->v.v0.cookie;
+    loop_run(loop);
+}
+
+    static void
+lcb_destroy_io_opts(struct lcb_io_opt_st *iops)
+{
+    rb_mt_loop *loop = iops->v.v0.cookie;
+    loop_destroy(loop);
+    free(iops);
+}
+
+    LIBCOUCHBASE_API lcb_error_t
+cb_create_ruby_mt_io_opts(int version, lcb_io_opt_t *io, void *arg)
+{
+    struct lcb_io_opt_st *ret;
+    rb_mt_loop *loop;
+    (void)arg;
+    if (version != 0) {
+        return LCB_PLUGIN_VERSION_MISMATCH;
+    }
+    ret = calloc(1, sizeof(*ret));
+    if (ret == NULL) {
+        free(ret);
+        return LCB_CLIENT_ENOMEM;
+    }
+
+    ret->version = 0;
+    ret->dlhandle = NULL;
+    ret->destructor = lcb_destroy_io_opts;
+    /* consider that struct isn't allocated by the library,
+     * `need_cleanup' flag might be set in lcb_create() */
+    ret->v.v0.need_cleanup = 0;
+    ret->v.v0.recv = cb_io_recv;
+    ret->v.v0.send = cb_io_send;
+    ret->v.v0.recvv = cb_io_recvv;
+    ret->v.v0.sendv = cb_io_sendv;
+    ret->v.v0.socket = cb_io_socket;
+    ret->v.v0.close = cb_io_close;
+    ret->v.v0.connect = cb_io_connect;
+    ret->v.v0.delete_event = lcb_io_delete_event;
+    ret->v.v0.destroy_event = lcb_io_destroy_event;
+    ret->v.v0.create_event = lcb_io_create_event;
+    ret->v.v0.update_event = lcb_io_update_event;
+
+    ret->v.v0.delete_timer = lcb_io_delete_timer;
+    ret->v.v0.destroy_timer = lcb_io_destroy_timer;
+    ret->v.v0.create_timer = lcb_io_create_timer;
+    ret->v.v0.update_timer = lcb_io_update_timer;
+
+    ret->v.v0.run_event_loop = lcb_io_run_event_loop;
+    ret->v.v0.stop_event_loop = lcb_io_stop_event_loop;
+
+    loop = loop_create();
+    if (loop == NULL) {
+        free(ret);
+        return LCB_CLIENT_ENOMEM;
+    }
+    ret->v.v0.cookie = loop;
+    *io = ret;
+    return LCB_SUCCESS;
+}
+#endif /* _WIN32 */