grpc 0.14.1 → 0.15.0

data/src/core/lib/iomgr/error.h

@@ -0,0 +1,192 @@
+/*
+ *
+ * Copyright 2016, Google Inc.
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ *
+ *     * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following disclaimer
+ * in the documentation and/or other materials provided with the
+ * distribution.
+ *     * Neither the name of Google Inc. nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ */
+
+#ifndef GRPC_CORE_LIB_IOMGR_ERROR_H
+#define GRPC_CORE_LIB_IOMGR_ERROR_H
+
+#include <stdbool.h>
+#include <stdint.h>
+
+#include <grpc/support/time.h>
+
+/// Opaque representation of an error.
+/// Errors are refcounted objects that represent the result of an operation.
+/// Ownership laws:
+///  if a grpc_error is returned by a function, the caller owns a ref to that
+///    instance
+///  if a grpc_error is passed to a grpc_closure callback function (functions
+///    with the signature:
+///      void (*f)(grpc_exec_ctx *exec_ctx, void *arg, grpc_error *error))
+///    then those functions do not automatically own a ref to error
+///  if a grpc_error is passed to *ANY OTHER FUNCTION* then that function takes
+///    ownership of the error
+/// Errors have:
+///  a set of ints, strings, and timestamps that describe the error
+///  always present are:
+///    GRPC_ERROR_STR_FILE, GRPC_ERROR_INT_FILE_LINE - source location the error
+///      was generated
+///    GRPC_ERROR_STR_DESCRIPTION - a human readable description of the error
+///    GRPC_ERROR_TIME_CREATED - a timestamp indicating when the error happened
+///  an error can also have children; these are other errors that are believed
+///    to have contributed to this one. By accumulating children, we can begin
+///    to root cause high level failures from low level failures, without having
+///    to derive execution paths from log lines
+typedef struct grpc_error grpc_error;
+
+typedef enum {
+  /// 'errno' from the operating system
+  GRPC_ERROR_INT_ERRNO,
+  /// __LINE__ from the call site creating the error
+  GRPC_ERROR_INT_FILE_LINE,
+  /// stream identifier: for errors that are associated with an individual
+  /// wire stream
+  GRPC_ERROR_INT_STREAM_ID,
+  /// grpc status code representing this error
+  GRPC_ERROR_INT_GRPC_STATUS,
+  /// offset into some binary blob (usually represented by
+  /// GRPC_ERROR_STR_RAW_BYTES) where the error occurred
+  GRPC_ERROR_INT_OFFSET,
+  /// context sensitive index associated with the error
+  GRPC_ERROR_INT_INDEX,
+  /// context sensitive size associated with the error
+  GRPC_ERROR_INT_SIZE,
+  /// http2 error code associated with the error (see the HTTP2 RFC)
+  GRPC_ERROR_INT_HTTP2_ERROR,
+  /// TSI status code associated with the error
+  GRPC_ERROR_INT_TSI_CODE,
+  /// grpc_security_status associated with the error
+  GRPC_ERROR_INT_SECURITY_STATUS,
+  /// WSAGetLastError() reported when this error occurred
+  GRPC_ERROR_INT_WSA_ERROR,
+  /// File descriptor associated with this error
+  GRPC_ERROR_INT_FD,
+  /// HTTP status (i.e. 404)
+  GRPC_ERROR_INT_HTTP_STATUS,
+} grpc_error_ints;
+
+typedef enum {
+  /// top-level textual description of this error
+  GRPC_ERROR_STR_DESCRIPTION,
+  /// source file in which this error occurred
+  GRPC_ERROR_STR_FILE,
+  /// operating system description of this error
+  GRPC_ERROR_STR_OS_ERROR,
+  /// syscall that generated this error
+  GRPC_ERROR_STR_SYSCALL,
+  /// peer that we were trying to communicate when this error occurred
+  GRPC_ERROR_STR_TARGET_ADDRESS,
+  /// grpc status message associated with this error
+  GRPC_ERROR_STR_GRPC_MESSAGE,
+  /// hex dump (or similar) with the data that generated this error
+  GRPC_ERROR_STR_RAW_BYTES,
+  /// tsi error string associated with this error
+  GRPC_ERROR_STR_TSI_ERROR,
+  /// filename that we were trying to read/write when this error occurred
+  GRPC_ERROR_STR_FILENAME,
+} grpc_error_strs;
+
+typedef enum {
+  /// timestamp of error creation
+  GRPC_ERROR_TIME_CREATED,
+} grpc_error_times;
+
+#define GRPC_ERROR_NONE ((grpc_error *)NULL)
+#define GRPC_ERROR_OOM ((grpc_error *)1)
+#define GRPC_ERROR_CANCELLED ((grpc_error *)2)
+
+const char *grpc_error_string(grpc_error *error);
+void grpc_error_free_string(const char *str);
+
+/// Create an error - but use GRPC_ERROR_CREATE instead
+grpc_error *grpc_error_create(const char *file, int line, const char *desc,
+                              grpc_error **referencing, size_t num_referencing);
+/// Create an error (this is the preferred way of generating an error that is
+///   not due to a system call - for system calls, use GRPC_OS_ERROR or
+///   GRPC_WSA_ERROR as appropriate)
+/// \a referencing is an array of num_referencing elements indicating one or
+/// more errors that are believed to have contributed to this one
+/// err = grpc_error_create(x, y, z, r, nr) is equivalent to:
+///   err = grpc_error_create(x, y, z, NULL, 0);
+///   for (i=0; i<nr; i++) err = grpc_error_add_child(err, r[i]);
+#define GRPC_ERROR_CREATE(desc) \
+  grpc_error_create(__FILE__, __LINE__, desc, NULL, 0)
+
+// Create an error that references some other errors. This function adds a
+// reference to each error in errs - it does not consume an existing reference
+#define GRPC_ERROR_CREATE_REFERENCING(desc, errs, count) \
+  grpc_error_create(__FILE__, __LINE__, desc, errs, count)
+
+//#define GRPC_ERROR_REFCOUNT_DEBUG
+#ifdef GRPC_ERROR_REFCOUNT_DEBUG
+grpc_error *grpc_error_ref(grpc_error *err, const char *file, int line,
+                           const char *func);
+void grpc_error_unref(grpc_error *err, const char *file, int line,
+                      const char *func);
+#define GRPC_ERROR_REF(err) grpc_error_ref(err, __FILE__, __LINE__, __func__)
+#define GRPC_ERROR_UNREF(err) \
+  grpc_error_unref(err, __FILE__, __LINE__, __func__)
+#else
+grpc_error *grpc_error_ref(grpc_error *err);
+void grpc_error_unref(grpc_error *err);
+#define GRPC_ERROR_REF(err) grpc_error_ref(err)
+#define GRPC_ERROR_UNREF(err) grpc_error_unref(err)
+#endif
+
+grpc_error *grpc_error_set_int(grpc_error *src, grpc_error_ints which,
+                               intptr_t value);
+bool grpc_error_get_int(grpc_error *error, grpc_error_ints which, intptr_t *p);
+grpc_error *grpc_error_set_time(grpc_error *src, grpc_error_times which,
+                                gpr_timespec value);
+grpc_error *grpc_error_set_str(grpc_error *src, grpc_error_strs which,
+                               const char *value);
+/// Add a child error: an error that is believed to have contributed to this
+/// error occurring. Allows root causing high level errors from lower level
+/// errors that contributed to them.
+grpc_error *grpc_error_add_child(grpc_error *src, grpc_error *child);
+grpc_error *grpc_os_error(const char *file, int line, int err,
+                          const char *call_name);
+/// create an error associated with errno!=0 (an 'operating system' error)
+#define GRPC_OS_ERROR(err, call_name) \
+  grpc_os_error(__FILE__, __LINE__, err, call_name)
+grpc_error *grpc_wsa_error(const char *file, int line, int err,
+                           const char *call_name);
+/// windows only: create an error associated with WSAGetLastError()!=0
+#define GRPC_WSA_ERROR(err, call_name) \
+  grpc_wsa_error(__FILE__, __LINE__, err, call_name)
+
+bool grpc_log_if_error(const char *what, grpc_error *error, const char *file,
+                       int line);
+#define GRPC_LOG_IF_ERROR(what, error) \
+  grpc_log_if_error((what), (error), __FILE__, __LINE__)
+
+#endif /* GRPC_CORE_LIB_IOMGR_ERROR_H */

data/src/core/lib/iomgr/ev_poll_and_epoll_posix.c

@@ -126,6 +126,9 @@ struct grpc_fd {
   grpc_closure *on_done_closure;
 
   grpc_iomgr_object iomgr_object;
+
+  /* The pollset that last noticed and notified that the fd is readable */
+  grpc_pollset *read_notifier_pollset;
 };
 
 /* Begin polling on an fd.
@@ -147,7 +150,8 @@ static uint32_t fd_begin_poll(grpc_fd *fd, grpc_pollset *pollset,
    if got_read or got_write are 1, also does the become_{readable,writable} as
    appropriate. */
 static void fd_end_poll(grpc_exec_ctx *exec_ctx, grpc_fd_watcher *rec,
-                        int got_read, int got_write);
+                        int got_read, int got_write,
+                        grpc_pollset *read_notifier_pollset);
 
 /* Return 1 if this fd is orphaned, 0 otherwise */
 static bool fd_is_orphaned(grpc_fd *fd);
@@ -217,9 +221,10 @@ struct grpc_pollset {
 struct grpc_pollset_vtable {
   void (*add_fd)(grpc_exec_ctx *exec_ctx, grpc_pollset *pollset,
                  struct grpc_fd *fd, int and_unlock_pollset);
-  void (*maybe_work_and_unlock)(grpc_exec_ctx *exec_ctx, grpc_pollset *pollset,
-                                grpc_pollset_worker *worker,
-                                gpr_timespec deadline, gpr_timespec now);
+  grpc_error *(*maybe_work_and_unlock)(grpc_exec_ctx *exec_ctx,
+                                       grpc_pollset *pollset,
+                                       grpc_pollset_worker *worker,
+                                       gpr_timespec deadline, gpr_timespec now);
   void (*finish_shutdown)(grpc_pollset *pollset);
   void (*destroy)(grpc_pollset *pollset);
 };
@@ -247,9 +252,9 @@ static int poll_deadline_to_millis_timeout(gpr_timespec deadline,
 #define GRPC_POLLSET_REEVALUATE_POLLING_ON_WAKEUP 2
 /* As per pollset_kick, with an extended set of flags (defined above)
    -- mostly for fd_posix's use. */
-static void pollset_kick_ext(grpc_pollset *p,
-                             grpc_pollset_worker *specific_worker,
-                             uint32_t flags);
+static grpc_error *pollset_kick_ext(grpc_pollset *p,
+                                    grpc_pollset_worker *specific_worker,
+                                    uint32_t flags) GRPC_MUST_USE_RESULT;
 
 /* turn a pollset into a multipoller: platform specific */
 typedef void (*platform_become_multipoller_type)(grpc_exec_ctx *exec_ctx,
@@ -342,6 +347,7 @@ static grpc_fd *alloc_fd(int fd) {
   r->on_done_closure = NULL;
   r->closed = 0;
   r->released = 0;
+  r->read_notifier_pollset = NULL;
   gpr_mu_unlock(&r->mu);
   return r;
 }
@@ -415,12 +421,13 @@ static bool fd_is_orphaned(grpc_fd *fd) {
   return (gpr_atm_acq_load(&fd->refst) & 1) == 0;
 }
 
-static void pollset_kick_locked(grpc_fd_watcher *watcher) {
+static grpc_error *pollset_kick_locked(grpc_fd_watcher *watcher) {
   gpr_mu_lock(&watcher->pollset->mu);
   GPR_ASSERT(watcher->worker);
-  pollset_kick_ext(watcher->pollset, watcher->worker,
-                   GRPC_POLLSET_REEVALUATE_POLLING_ON_WAKEUP);
+  grpc_error *err = pollset_kick_ext(watcher->pollset, watcher->worker,
+                                     GRPC_POLLSET_REEVALUATE_POLLING_ON_WAKEUP);
   gpr_mu_unlock(&watcher->pollset->mu);
+  return err;
 }
 
 static void maybe_wake_one_watcher_locked(grpc_fd *fd) {
@@ -459,7 +466,7 @@ static void close_fd_locked(grpc_exec_ctx *exec_ctx, grpc_fd *fd) {
   } else {
     remove_fd_from_all_epoll_sets(fd->fd);
   }
-  grpc_exec_ctx_enqueue(exec_ctx, fd->on_done_closure, true, NULL);
+  grpc_exec_ctx_sched(exec_ctx, fd->on_done_closure, GRPC_ERROR_NONE, NULL);
 }
 
 static int fd_wrapped_fd(grpc_fd *fd) {
@@ -508,15 +515,27 @@ static void fd_ref(grpc_fd *fd) { ref_by(fd, 2); }
 static void fd_unref(grpc_fd *fd) { unref_by(fd, 2); }
 #endif
 
+static grpc_error *fd_shutdown_error(bool shutdown) {
+  if (!shutdown) {
+    return GRPC_ERROR_NONE;
+  } else {
+    return GRPC_ERROR_CREATE("FD shutdown");
+  }
+}
+
 static void notify_on_locked(grpc_exec_ctx *exec_ctx, grpc_fd *fd,
                              grpc_closure **st, grpc_closure *closure) {
-  if (*st == CLOSURE_NOT_READY) {
+  if (fd->shutdown) {
+    grpc_exec_ctx_sched(exec_ctx, closure, GRPC_ERROR_CREATE("FD shutdown"),
+                        NULL);
+  } else if (*st == CLOSURE_NOT_READY) {
     /* not ready ==> switch to a waiting state by setting the closure */
     *st = closure;
   } else if (*st == CLOSURE_READY) {
     /* already ready ==> queue the closure to run immediately */
     *st = CLOSURE_NOT_READY;
-    grpc_exec_ctx_enqueue(exec_ctx, closure, !fd->shutdown, NULL);
+    grpc_exec_ctx_sched(exec_ctx, closure, fd_shutdown_error(fd->shutdown),
+                        NULL);
     maybe_wake_one_watcher_locked(fd);
   } else {
     /* upcallptr was set to a different closure.  This is an error! */
@@ -539,19 +558,35 @@ static int set_ready_locked(grpc_exec_ctx *exec_ctx, grpc_fd *fd,
     return 0;
   } else {
     /* waiting ==> queue closure */
-    grpc_exec_ctx_enqueue(exec_ctx, *st, !fd->shutdown, NULL);
+    grpc_exec_ctx_sched(exec_ctx, *st, fd_shutdown_error(fd->shutdown), NULL);
     *st = CLOSURE_NOT_READY;
     return 1;
   }
 }
 
+static void set_read_notifier_pollset_locked(
+    grpc_exec_ctx *exec_ctx, grpc_fd *fd, grpc_pollset *read_notifier_pollset) {
+  fd->read_notifier_pollset = read_notifier_pollset;
+}
+
 static void fd_shutdown(grpc_exec_ctx *exec_ctx, grpc_fd *fd) {
   gpr_mu_lock(&fd->mu);
-  GPR_ASSERT(!fd->shutdown);
-  fd->shutdown = 1;
-  set_ready_locked(exec_ctx, fd, &fd->read_closure);
-  set_ready_locked(exec_ctx, fd, &fd->write_closure);
+  /* only shutdown once */
+  if (!fd->shutdown) {
+    fd->shutdown = 1;
+    /* signal read/write closed to OS so that future operations fail */
+    shutdown(fd->fd, SHUT_RDWR);
+    set_ready_locked(exec_ctx, fd, &fd->read_closure);
+    set_ready_locked(exec_ctx, fd, &fd->write_closure);
+  }
+  gpr_mu_unlock(&fd->mu);
+}
+
+static bool fd_is_shutdown(grpc_fd *fd) {
+  gpr_mu_lock(&fd->mu);
+  bool r = fd->shutdown;
   gpr_mu_unlock(&fd->mu);
+  return r;
 }
 
 static void fd_notify_on_read(grpc_exec_ctx *exec_ctx, grpc_fd *fd,
@@ -568,6 +603,18 @@ static void fd_notify_on_write(grpc_exec_ctx *exec_ctx, grpc_fd *fd,
   gpr_mu_unlock(&fd->mu);
 }
 
+/* Return the read-notifier pollset */
+static grpc_pollset *fd_get_read_notifier_pollset(grpc_exec_ctx *exec_ctx,
+                                                  grpc_fd *fd) {
+  grpc_pollset *notifier = NULL;
+
+  gpr_mu_lock(&fd->mu);
+  notifier = fd->read_notifier_pollset;
+  gpr_mu_unlock(&fd->mu);
+
+  return notifier;
+}
+
 static uint32_t fd_begin_poll(grpc_fd *fd, grpc_pollset *pollset,
                               grpc_pollset_worker *worker, uint32_t read_mask,
                               uint32_t write_mask, grpc_fd_watcher *watcher) {
@@ -620,7 +667,8 @@ static uint32_t fd_begin_poll(grpc_fd *fd, grpc_pollset *pollset,
 }
 
 static void fd_end_poll(grpc_exec_ctx *exec_ctx, grpc_fd_watcher *watcher,
-                        int got_read, int got_write) {
+                        int got_read, int got_write,
+                        grpc_pollset *read_notifier_pollset) {
   int was_polling = 0;
   int kick = 0;
   grpc_fd *fd = watcher->fd;
@@ -656,6 +704,10 @@ static void fd_end_poll(grpc_exec_ctx *exec_ctx, grpc_fd_watcher *watcher,
     if (set_ready_locked(exec_ctx, fd, &fd->read_closure)) {
       kick = 1;
     }
+
+    if (read_notifier_pollset != NULL) {
+      set_read_notifier_pollset_locked(exec_ctx, fd, read_notifier_pollset);
+    }
   }
   if (got_write) {
     if (set_ready_locked(exec_ctx, fd, &fd->write_closure)) {
@@ -717,10 +769,19 @@ static void push_front_worker(grpc_pollset *p, grpc_pollset_worker *worker) {
   worker->prev->next = worker->next->prev = worker;
 }
 
-static void pollset_kick_ext(grpc_pollset *p,
-                             grpc_pollset_worker *specific_worker,
-                             uint32_t flags) {
+static void kick_append_error(grpc_error **composite, grpc_error *error) {
+  if (error == GRPC_ERROR_NONE) return;
+  if (*composite == GRPC_ERROR_NONE) {
+    *composite = GRPC_ERROR_CREATE("Kick Failure");
+  }
+  *composite = grpc_error_add_child(*composite, error);
+}
+
+static grpc_error *pollset_kick_ext(grpc_pollset *p,
+                                    grpc_pollset_worker *specific_worker,
+                                    uint32_t flags) {
   GPR_TIMER_BEGIN("pollset_kick_ext", 0);
+  grpc_error *error = GRPC_ERROR_NONE;
 
   /* pollset->mu already held */
   if (specific_worker != NULL) {
@@ -730,25 +791,28 @@ static void pollset_kick_ext(grpc_pollset *p,
       for (specific_worker = p->root_worker.next;
            specific_worker != &p->root_worker;
            specific_worker = specific_worker->next) {
-        grpc_wakeup_fd_wakeup(&specific_worker->wakeup_fd->fd);
+        kick_append_error(
+            &error, grpc_wakeup_fd_wakeup(&specific_worker->wakeup_fd->fd));
       }
-      p->kicked_without_pollers = 1;
+      p->kicked_without_pollers = true;
       GPR_TIMER_END("pollset_kick_ext.broadcast", 0);
     } else if (gpr_tls_get(&g_current_thread_worker) !=
                (intptr_t)specific_worker) {
       GPR_TIMER_MARK("different_thread_worker", 0);
       if ((flags & GRPC_POLLSET_REEVALUATE_POLLING_ON_WAKEUP) != 0) {
-        specific_worker->reevaluate_polling_on_wakeup = 1;
+        specific_worker->reevaluate_polling_on_wakeup = true;
       }
-      specific_worker->kicked_specifically = 1;
-      grpc_wakeup_fd_wakeup(&specific_worker->wakeup_fd->fd);
+      specific_worker->kicked_specifically = true;
+      kick_append_error(&error,
+                        grpc_wakeup_fd_wakeup(&specific_worker->wakeup_fd->fd));
     } else if ((flags & GRPC_POLLSET_CAN_KICK_SELF) != 0) {
       GPR_TIMER_MARK("kick_yoself", 0);
       if ((flags & GRPC_POLLSET_REEVALUATE_POLLING_ON_WAKEUP) != 0) {
-        specific_worker->reevaluate_polling_on_wakeup = 1;
+        specific_worker->reevaluate_polling_on_wakeup = true;
       }
-      specific_worker->kicked_specifically = 1;
-      grpc_wakeup_fd_wakeup(&specific_worker->wakeup_fd->fd);
+      specific_worker->kicked_specifically = true;
+      kick_append_error(&error,
+                        grpc_wakeup_fd_wakeup(&specific_worker->wakeup_fd->fd));
     }
   } else if (gpr_tls_get(&g_current_thread_poller) != (intptr_t)p) {
     GPR_ASSERT((flags & GRPC_POLLSET_REEVALUATE_POLLING_ON_WAKEUP) == 0);
@@ -769,39 +833,41 @@ static void pollset_kick_ext(grpc_pollset *p,
       if (specific_worker != NULL) {
         GPR_TIMER_MARK("finally_kick", 0);
         push_back_worker(p, specific_worker);
-        grpc_wakeup_fd_wakeup(&specific_worker->wakeup_fd->fd);
+        kick_append_error(
+            &error, grpc_wakeup_fd_wakeup(&specific_worker->wakeup_fd->fd));
       }
     } else {
       GPR_TIMER_MARK("kicked_no_pollers", 0);
-      p->kicked_without_pollers = 1;
+      p->kicked_without_pollers = true;
     }
   }
 
   GPR_TIMER_END("pollset_kick_ext", 0);
+  return error;
 }
 
-static void pollset_kick(grpc_pollset *p,
-                         grpc_pollset_worker *specific_worker) {
-  pollset_kick_ext(p, specific_worker, 0);
+static grpc_error *pollset_kick(grpc_pollset *p,
+                                grpc_pollset_worker *specific_worker) {
+  return pollset_kick_ext(p, specific_worker, 0);
 }
 
 /* global state management */
 
-static void pollset_global_init(void) {
+static grpc_error *pollset_global_init(void) {
   gpr_tls_init(&g_current_thread_poller);
   gpr_tls_init(&g_current_thread_worker);
-  grpc_wakeup_fd_global_init();
-  grpc_wakeup_fd_init(&grpc_global_wakeup_fd);
+  return grpc_wakeup_fd_init(&grpc_global_wakeup_fd);
 }
 
 static void pollset_global_shutdown(void) {
   grpc_wakeup_fd_destroy(&grpc_global_wakeup_fd);
   gpr_tls_destroy(&g_current_thread_poller);
   gpr_tls_destroy(&g_current_thread_worker);
-  grpc_wakeup_fd_global_destroy();
 }
 
-static void kick_poller(void) { grpc_wakeup_fd_wakeup(&grpc_global_wakeup_fd); }
+static grpc_error *kick_poller(void) {
+  return grpc_wakeup_fd_wakeup(&grpc_global_wakeup_fd);
+}
 
 /* main interface */
 
@@ -864,14 +930,23 @@ static void pollset_add_fd(grpc_exec_ctx *exec_ctx, grpc_pollset *pollset,
 static void finish_shutdown(grpc_exec_ctx *exec_ctx, grpc_pollset *pollset) {
   GPR_ASSERT(grpc_closure_list_empty(pollset->idle_jobs));
   pollset->vtable->finish_shutdown(pollset);
-  grpc_exec_ctx_enqueue(exec_ctx, pollset->shutdown_done, true, NULL);
+  grpc_exec_ctx_sched(exec_ctx, pollset->shutdown_done, GRPC_ERROR_NONE, NULL);
+}
+
+static void work_combine_error(grpc_error **composite, grpc_error *error) {
+  if (error == GRPC_ERROR_NONE) return;
+  if (*composite == GRPC_ERROR_NONE) {
+    *composite = GRPC_ERROR_CREATE("pollset_work");
+  }
+  *composite = grpc_error_add_child(*composite, error);
 }
 
-static void pollset_work(grpc_exec_ctx *exec_ctx, grpc_pollset *pollset,
-                         grpc_pollset_worker **worker_hdl, gpr_timespec now,
-                         gpr_timespec deadline) {
+static grpc_error *pollset_work(grpc_exec_ctx *exec_ctx, grpc_pollset *pollset,
+                                grpc_pollset_worker **worker_hdl,
+                                gpr_timespec now, gpr_timespec deadline) {
   grpc_pollset_worker worker;
   *worker_hdl = &worker;
+  grpc_error *error = GRPC_ERROR_NONE;
 
   /* pollset->mu already held */
   int added_worker = 0;
@@ -887,7 +962,10 @@ static void pollset_work(grpc_exec_ctx *exec_ctx, grpc_pollset *pollset,
     pollset->local_wakeup_cache = worker.wakeup_fd->next;
   } else {
     worker.wakeup_fd = gpr_malloc(sizeof(*worker.wakeup_fd));
-    grpc_wakeup_fd_init(&worker.wakeup_fd->fd);
+    error = grpc_wakeup_fd_init(&worker.wakeup_fd->fd);
+    if (error != GRPC_ERROR_NONE) {
+      return error;
+    }
   }
   worker.kicked_specifically = 0;
   /* If there's work waiting for the pollset to be idle, and the
@@ -924,8 +1002,9 @@ static void pollset_work(grpc_exec_ctx *exec_ctx, grpc_pollset *pollset,
       }
       gpr_tls_set(&g_current_thread_poller, (intptr_t)pollset);
       GPR_TIMER_BEGIN("maybe_work_and_unlock", 0);
-      pollset->vtable->maybe_work_and_unlock(exec_ctx, pollset, &worker,
-                                             deadline, now);
+      work_combine_error(&error,
+                         pollset->vtable->maybe_work_and_unlock(
+                             exec_ctx, pollset, &worker, deadline, now));
       GPR_TIMER_END("maybe_work_and_unlock", 0);
       locked = 0;
       gpr_tls_set(&g_current_thread_poller, 0);
@@ -987,6 +1066,7 @@ static void pollset_work(grpc_exec_ctx *exec_ctx, grpc_pollset *pollset,
   }
   *worker_hdl = NULL;
   GPR_TIMER_END("pollset_work", 0);
+  return error;
 }
 
 static void pollset_shutdown(grpc_exec_ctx *exec_ctx, grpc_pollset *pollset,
@@ -1035,7 +1115,7 @@ typedef struct grpc_unary_promote_args {
 } grpc_unary_promote_args;
 
 static void basic_do_promote(grpc_exec_ctx *exec_ctx, void *args,
-                             bool success) {
+                             grpc_error *error) {
   grpc_unary_promote_args *up_args = args;
   const grpc_pollset_vtable *original_vtable = up_args->original_vtable;
   grpc_pollset *pollset = up_args->pollset;
@@ -1137,7 +1217,8 @@ static void basic_pollset_add_fd(grpc_exec_ctx *exec_ctx, grpc_pollset *pollset,
   up_args->promotion_closure.cb = basic_do_promote;
   up_args->promotion_closure.cb_arg = up_args;
 
-  grpc_closure_list_add(&pollset->idle_jobs, &up_args->promotion_closure, 1);
+  grpc_closure_list_append(&pollset->idle_jobs, &up_args->promotion_closure,
+                           GRPC_ERROR_NONE);
   pollset_kick(pollset, GRPC_POLLSET_KICK_BROADCAST);
 
 exit:
@@ -1146,11 +1227,9 @@ exit:
   }
 }
 
-static void basic_pollset_maybe_work_and_unlock(grpc_exec_ctx *exec_ctx,
-                                                grpc_pollset *pollset,
-                                                grpc_pollset_worker *worker,
-                                                gpr_timespec deadline,
-                                                gpr_timespec now) {
+static grpc_error *basic_pollset_maybe_work_and_unlock(
+    grpc_exec_ctx *exec_ctx, grpc_pollset *pollset, grpc_pollset_worker *worker,
+    gpr_timespec deadline, gpr_timespec now) {
 #define POLLOUT_CHECK (POLLOUT | POLLHUP | POLLERR)
 #define POLLIN_CHECK (POLLIN | POLLHUP | POLLERR)
 
@@ -1160,6 +1239,7 @@ static void basic_pollset_maybe_work_and_unlock(grpc_exec_ctx *exec_ctx,
   int timeout;
   int r;
   nfds_t nfds;
+  grpc_error *error = GRPC_ERROR_NONE;
 
   fd = pollset->data.ptr;
   if (fd && fd_is_orphaned(fd)) {
@@ -1200,33 +1280,37 @@ static void basic_pollset_maybe_work_and_unlock(grpc_exec_ctx *exec_ctx,
 
   if (r < 0) {
     if (errno != EINTR) {
-      gpr_log(GPR_ERROR, "poll() failed: %s", strerror(errno));
+      work_combine_error(&error, GRPC_OS_ERROR(errno, "poll"));
     }
     if (fd) {
-      fd_end_poll(exec_ctx, &fd_watcher, 0, 0);
+      fd_end_poll(exec_ctx, &fd_watcher, 0, 0, NULL);
     }
   } else if (r == 0) {
     if (fd) {
-      fd_end_poll(exec_ctx, &fd_watcher, 0, 0);
+      fd_end_poll(exec_ctx, &fd_watcher, 0, 0, NULL);
     }
   } else {
     if (pfd[0].revents & POLLIN_CHECK) {
-      grpc_wakeup_fd_consume_wakeup(&grpc_global_wakeup_fd);
+      work_combine_error(&error,
+                         grpc_wakeup_fd_consume_wakeup(&grpc_global_wakeup_fd));
     }
     if (pfd[1].revents & POLLIN_CHECK) {
-      grpc_wakeup_fd_consume_wakeup(&worker->wakeup_fd->fd);
+      work_combine_error(&error,
+                         grpc_wakeup_fd_consume_wakeup(&worker->wakeup_fd->fd));
     }
     if (nfds > 2) {
       fd_end_poll(exec_ctx, &fd_watcher, pfd[2].revents & POLLIN_CHECK,
-                  pfd[2].revents & POLLOUT_CHECK);
+                  pfd[2].revents & POLLOUT_CHECK, pollset);
     } else if (fd) {
-      fd_end_poll(exec_ctx, &fd_watcher, 0, 0);
+      fd_end_poll(exec_ctx, &fd_watcher, 0, 0, NULL);
     }
   }
 
   if (fd) {
     GRPC_FD_UNREF(fd, "basicpoll_begin");
   }
+
+  return error;
 }
 
 static void basic_pollset_destroy(grpc_pollset *pollset) {
@@ -1287,7 +1371,7 @@ exit:
   }
 }
 
-static void multipoll_with_poll_pollset_maybe_work_and_unlock(
+static grpc_error *multipoll_with_poll_pollset_maybe_work_and_unlock(
     grpc_exec_ctx *exec_ctx, grpc_pollset *pollset, grpc_pollset_worker *worker,
     gpr_timespec deadline, gpr_timespec now) {
 #define POLLOUT_CHECK (POLLOUT | POLLHUP | POLLERR)
@@ -1301,6 +1385,7 @@ static void multipoll_with_poll_pollset_maybe_work_and_unlock(
   /* TODO(ctiller): inline some elements to avoid an allocation */
   grpc_fd_watcher *watchers;
   struct pollfd *pfds;
+  grpc_error *error = GRPC_ERROR_NONE;
 
   h = pollset->data.ptr;
   timeout = poll_deadline_to_millis_timeout(deadline, now);
@@ -1353,34 +1438,38 @@ static void multipoll_with_poll_pollset_maybe_work_and_unlock(
 
   if (r < 0) {
     if (errno != EINTR) {
-      gpr_log(GPR_ERROR, "poll() failed: %s", strerror(errno));
+      work_combine_error(&error, GRPC_OS_ERROR(errno, "poll"));
     }
     for (i = 2; i < pfd_count; i++) {
-      fd_end_poll(exec_ctx, &watchers[i], 0, 0);
+      fd_end_poll(exec_ctx, &watchers[i], 0, 0, NULL);
     }
   } else if (r == 0) {
     for (i = 2; i < pfd_count; i++) {
-      fd_end_poll(exec_ctx, &watchers[i], 0, 0);
+      fd_end_poll(exec_ctx, &watchers[i], 0, 0, NULL);
     }
   } else {
     if (pfds[0].revents & POLLIN_CHECK) {
-      grpc_wakeup_fd_consume_wakeup(&grpc_global_wakeup_fd);
+      work_combine_error(&error,
+                         grpc_wakeup_fd_consume_wakeup(&grpc_global_wakeup_fd));
     }
     if (pfds[1].revents & POLLIN_CHECK) {
-      grpc_wakeup_fd_consume_wakeup(&worker->wakeup_fd->fd);
+      work_combine_error(&error,
+                         grpc_wakeup_fd_consume_wakeup(&worker->wakeup_fd->fd));
     }
     for (i = 2; i < pfd_count; i++) {
       if (watchers[i].fd == NULL) {
-        fd_end_poll(exec_ctx, &watchers[i], 0, 0);
+        fd_end_poll(exec_ctx, &watchers[i], 0, 0, NULL);
         continue;
       }
       fd_end_poll(exec_ctx, &watchers[i], pfds[i].revents & POLLIN_CHECK,
-                  pfds[i].revents & POLLOUT_CHECK);
+                  pfds[i].revents & POLLOUT_CHECK, pollset);
     }
   }
 
   gpr_free(pfds);
   gpr_free(watchers);
+
+  return error;
 }
 
 static void multipoll_with_poll_pollset_finish_shutdown(grpc_pollset *pollset) {
@@ -1451,20 +1540,31 @@ static void poll_become_multipoller(grpc_exec_ctx *exec_ctx,
 #include "src/core/lib/profiling/timers.h"
 #include "src/core/lib/support/block_annotate.h"
 
-static void set_ready(grpc_exec_ctx *exec_ctx, grpc_fd *fd, grpc_closure **st) {
+static void set_ready(grpc_exec_ctx *exec_ctx, grpc_fd *fd, grpc_closure **st,
+                      grpc_pollset *read_notifier_pollset) {
   /* only one set_ready can be active at once (but there may be a racing
      notify_on) */
   gpr_mu_lock(&fd->mu);
   set_ready_locked(exec_ctx, fd, st);
+
+  /* A non-NULL read_notifier_pollset means that the fd is readable. */
+  if (read_notifier_pollset != NULL) {
+    /* Note: Since the fd might be a part of multiple pollsets, this might be
+     * called multiple times (for each time the fd becomes readable) and it is
+     * okay to set the fd's read-notifier pollset to anyone of these pollsets */
+    set_read_notifier_pollset_locked(exec_ctx, fd, read_notifier_pollset);
+  }
+
   gpr_mu_unlock(&fd->mu);
 }
 
-static void fd_become_readable(grpc_exec_ctx *exec_ctx, grpc_fd *fd) {
-  set_ready(exec_ctx, fd, &fd->read_closure);
+static void fd_become_readable(grpc_exec_ctx *exec_ctx, grpc_fd *fd,
+                               grpc_pollset *notifier_pollset) {
+  set_ready(exec_ctx, fd, &fd->read_closure, notifier_pollset);
 }
 
 static void fd_become_writable(grpc_exec_ctx *exec_ctx, grpc_fd *fd) {
-  set_ready(exec_ctx, fd, &fd->write_closure);
+  set_ready(exec_ctx, fd, &fd->write_closure, NULL);
 }
 
 struct epoll_fd_list {
@@ -1556,11 +1656,11 @@ static void finally_add_fd(grpc_exec_ctx *exec_ctx, grpc_pollset *pollset,
       }
     }
   }
-  fd_end_poll(exec_ctx, &watcher, 0, 0);
+  fd_end_poll(exec_ctx, &watcher, 0, 0, NULL);
 }
 
 static void perform_delayed_add(grpc_exec_ctx *exec_ctx, void *arg,
-                                bool iomgr_status) {
+                                grpc_error *error) {
   delayed_add *da = arg;
 
   if (!fd_is_orphaned(da->fd)) {
@@ -1573,7 +1673,8 @@ static void perform_delayed_add(grpc_exec_ctx *exec_ctx, void *arg,
     /* We don't care about this pollset anymore. */
     if (da->pollset->in_flight_cbs == 0 && !da->pollset->called_shutdown) {
       da->pollset->called_shutdown = 1;
-      grpc_exec_ctx_enqueue(exec_ctx, da->pollset->shutdown_done, true, NULL);
+      grpc_exec_ctx_sched(exec_ctx, da->pollset->shutdown_done, GRPC_ERROR_NONE,
+                          NULL);
     }
   }
   gpr_mu_unlock(&da->pollset->mu);
@@ -1597,14 +1698,14 @@ static void multipoll_with_epoll_pollset_add_fd(grpc_exec_ctx *exec_ctx,
     GRPC_FD_REF(fd, "delayed_add");
     grpc_closure_init(&da->closure, perform_delayed_add, da);
     pollset->in_flight_cbs++;
-    grpc_exec_ctx_enqueue(exec_ctx, &da->closure, true, NULL);
+    grpc_exec_ctx_sched(exec_ctx, &da->closure, GRPC_ERROR_NONE, NULL);
   }
 }
 
 /* TODO(klempner): We probably want to turn this down a bit */
 #define GRPC_EPOLL_MAX_EVENTS 1000
 
-static void multipoll_with_epoll_pollset_maybe_work_and_unlock(
+static grpc_error *multipoll_with_epoll_pollset_maybe_work_and_unlock(
     grpc_exec_ctx *exec_ctx, grpc_pollset *pollset, grpc_pollset_worker *worker,
     gpr_timespec deadline, gpr_timespec now) {
   struct epoll_event ep_ev[GRPC_EPOLL_MAX_EVENTS];
@@ -1613,6 +1714,7 @@ static void multipoll_with_epoll_pollset_maybe_work_and_unlock(
   epoll_hdr *h = pollset->data.ptr;
   int timeout_ms;
   struct pollfd pfds[2];
+  grpc_error *error = GRPC_ERROR_NONE;
 
   /* If you want to ignore epoll's ability to sanely handle parallel pollers,
    * for a more apples-to-apples performance comparison with poll, add a
@@ -1641,13 +1743,14 @@ static void multipoll_with_epoll_pollset_maybe_work_and_unlock(
 
   if (poll_rv < 0) {
     if (errno != EINTR) {
-      gpr_log(GPR_ERROR, "poll() failed: %s", strerror(errno));
+      work_combine_error(&error, GRPC_OS_ERROR(errno, "poll"));
     }
   } else if (poll_rv == 0) {
     /* do nothing */
   } else {
     if (pfds[0].revents) {
-      grpc_wakeup_fd_consume_wakeup(&worker->wakeup_fd->fd);
+      work_combine_error(&error,
+                         grpc_wakeup_fd_consume_wakeup(&worker->wakeup_fd->fd));
     }
     if (pfds[1].revents) {
       do {
@@ -1655,7 +1758,7 @@ static void multipoll_with_epoll_pollset_maybe_work_and_unlock(
         ep_rv = epoll_wait(h->epoll_fd, ep_ev, GRPC_EPOLL_MAX_EVENTS, 0);
         if (ep_rv < 0) {
           if (errno != EINTR) {
-            gpr_log(GPR_ERROR, "epoll_wait() failed: %s", strerror(errno));
+            work_combine_error(&error, GRPC_OS_ERROR(errno, "epoll_wait"));
           }
         } else {
           int i;
@@ -1667,10 +1770,11 @@ static void multipoll_with_epoll_pollset_maybe_work_and_unlock(
             int read_ev = ep_ev[i].events & (EPOLLIN | EPOLLPRI);
             int write_ev = ep_ev[i].events & EPOLLOUT;
             if (fd == NULL) {
-              grpc_wakeup_fd_consume_wakeup(&grpc_global_wakeup_fd);
+              work_combine_error(&error, grpc_wakeup_fd_consume_wakeup(
+                                             &grpc_global_wakeup_fd));
             } else {
               if (read_ev || cancel) {
-                fd_become_readable(exec_ctx, fd);
+                fd_become_readable(exec_ctx, fd, pollset);
               }
               if (write_ev || cancel) {
                 fd_become_writable(exec_ctx, fd);
@@ -1681,6 +1785,7 @@ static void multipoll_with_epoll_pollset_maybe_work_and_unlock(
       } while (ep_rv == GRPC_EPOLL_MAX_EVENTS);
     }
   }
+  return error;
 }
 
 static void multipoll_with_epoll_pollset_finish_shutdown(
@@ -1897,8 +2002,10 @@ static const grpc_event_engine_vtable vtable = {
     .fd_wrapped_fd = fd_wrapped_fd,
     .fd_orphan = fd_orphan,
     .fd_shutdown = fd_shutdown,
+    .fd_is_shutdown = fd_is_shutdown,
     .fd_notify_on_read = fd_notify_on_read,
     .fd_notify_on_write = fd_notify_on_write,
+    .fd_get_read_notifier_pollset = fd_get_read_notifier_pollset,
 
     .pollset_init = pollset_init,
     .pollset_shutdown = pollset_shutdown,