1# Polling Engines 2 3_Author: Sree Kuchibhotla (@sreecha) - Sep 2018_ 4 5 6## Why do we need a 'polling engine' ? 7 8Polling engine component was created for the following reasons: 9 10- gRPC code deals with a bunch of file descriptors on which events like descriptor being readable/writable/error have to be monitored 11- gRPC code knows the actions to perform when such events happen 12 - For example: 13 - `grpc_endpoint` code calls `recvmsg` call when the fd is readable and `sendmsg` call when the fd is writable 14 - ` tcp_client` connect code issues async `connect` and finishes creating the client once the fd is writable (i.e when the `connect` actually finished) 15- gRPC needed some component that can "efficiently" do the above operations __using the threads provided by the applications (i.e., not create any new threads)__. Also by "efficiently" we mean optimized for latency and throughput 16 17 18## Polling Engine Implementations in gRPC 19There are multiple polling engine implementations depending on the OS and the OS version. Fortunately all of them expose the same interface 20 21- Linux: 22 23 - **`epollex`** (default but requires kernel version >= 4.5), 24 - `epoll1` (If `epollex` is not available and glibc version >= 2.9) 25 - `poll` (If kernel does not have epoll support) 26- Mac: **`poll`** (default) 27- Windows: (no name) 28- One-off polling engines: 29 - NodeJS : `libuv` polling engine implementation (requires different compile `#define`s) 30 31## Polling Engine Interface 32 33### Opaque Structures exposed by the polling engine 34The following are the **Opaque** structures exposed by Polling Engine interface (NOTE: Different polling engine implementations have different definitions of these structures) 35 36- **grpc_fd:** Structure representing a file descriptor 37- **grpc_pollset:** A set of one or more grpc_fds that are ‘polled’ for readable/writable/error events. One grpc_fd can be in multiple `grpc_pollset`s 38- **grpc_pollset_worker:** Structure representing a ‘polling thread’ - more specifically, the thread that calls `grpc_pollset_work()` API 39- **grpc_pollset_set:** A group of `grpc_fd`s, `grpc_pollset`s and `grpc_pollset_set`s (yes, a `grpc_pollset_set` can contain other `grpc_pollset_set`s) 40 41### Polling engine API 42 43#### grpc_fd 44- **grpc\_fd\_notify\_on\_[read|write|error]** 45 - Signature: `grpc_fd_notify_on_(grpc_fd* fd, grpc_closure* closure)` 46 - Register a [closure](https://github.com/grpc/grpc/blob/v1.15.1/src/core/lib/iomgr/closure.h#L67) to be called when the fd becomes readable/writable or has an error (In grpc parlance, we refer to this act as “arming the fd”) 47 - The closure is called exactly once per event. I.e once the fd becomes readable (or writable or error), the closure is fired and the fd is ‘unarmed’. To be notified again, the fd has to be armed again. 48 49- **grpc_fd_shutdown** 50 - Signature: `grpc_fd_shutdown(grpc_fd* fd)` 51 - Any current (or future) closures registered for readable/writable/error events are scheduled immediately with an error 52 53- **grpc_fd_orphan** 54 - Signature: `grpc_fd_orphan(grpc_fd* fd, grpc_closure* on_done, int* release_fd, char* reason)` 55 - Release the `grpc_fd` structure and call `on_done` closure when the operation is complete 56 - If `release_fd` is set to `nullptr`, then `close()` the underlying fd as well. If not, put the underlying fd in `release_fd` (and do not call `close()`) 57 - `release_fd` set to non-null in cases where the underlying fd is NOT owned by grpc core (like for example the fds used by C-Ares DNS resolver ) 58 59#### grpc_pollset 60 61- **grpc_pollset_add_fd** 62 - Signature: `grpc_pollset_add_fd(grpc_pollset* ps, grpc_fd *fd)` 63 - Add fd to pollset 64 > **NOTE**: There is no `grpc_pollset_remove_fd`. This is because calling `grpc_fd_orphan()` will effectively remove the fd from all the pollsets it’s a part of 65 66- **grpc_pollset_work** 67 - Signature: `grpc_pollset_work(grpc_pollset* ps, grpc_pollset_worker** worker, grpc_millis deadline)` 68 > **NOTE**: `grpc_pollset_work()` requires the pollset mutex to be locked before calling it. Shortly after calling `grpc_pollset_work()`, the function populates the `*worker` pointer (among other things) and releases the mutex. Once `grpc_pollset_work()` returns, the `*worker` pointer is **invalid** and should not be used anymore. See the code in `completion_queue.cc` to see how this is used. 69 - Poll the fds in the pollset for events AND return when ANY of the following is true: 70 - Deadline expired 71 - Some fds in the pollset were found to be readable/writable/error and those associated closures were ‘scheduled’ (but not necessarily executed) 72 - worker is “kicked” (see `grpc_pollset_kick` for more details) 73 74- **grpc_pollset_kick** 75 - Signature: `grpc_pollset_kick(grpc_pollset* ps, grpc_pollset_worker* worker)` 76 - “Kick the worker” i.e Force the worker to return from grpc_pollset_work() 77 - If `worker == nullptr`, kick ANY worker active on that pollset 78 79#### grpc_pollset_set 80 81- **grpc\_pollset\_set\_[add|del]\_fd** 82 - Signature: `grpc_pollset_set_[add|del]_fd(grpc_pollset_set* pss, grpc_fd *fd)` 83 - Add/Remove fd to the `grpc_pollset_set` 84 85- **grpc\_pollset\_set_[add|del]\_pollset** 86 - Signature: `grpc_pollset_set_[add|del]_pollset(grpc_pollset_set* pss, grpc_pollset* ps)` 87 - What does adding a pollset to a pollset_set mean ? 88 - It means that calling `grpc_pollset_work()` on the pollset will also poll all the fds in the pollset_set i.e semantically, it is similar to adding all the fds inside pollset_set to the pollset. 89 - This guarantee is no longer true once the pollset is removed from the pollset_set 90 91- **grpc\_pollset\_set_[add|del]\_pollset\_set** 92 - Signature: `grpc_pollset_set_[add|del]_pollset_set(grpc_pollset_set* bag, grpc_pollset_set* item)` 93 - Semantically, this is similar to adding all the fds in the ‘bag’ pollset_set to the ‘item’ pollset_set 94 95 96#### Recap: 97 98__Relation between grpc_pollset_worker, grpc_pollset and grpc_fd:__ 99 100 101 102__grpc_pollset_set__ 103 104 105 106 107## Polling Engine Implementations 108 109### epoll1 110 111 112 113Code at `src/core/lib/iomgr/ev_epoll1_posix.cc` 114 115- The logic to choose a designated poller is quite complicated. Pollsets are internally sharded into what are called `pollset_neighborhood` (a structure internal to `epoll1` polling engine implementation). `grpc_pollset_workers` that call `grpc_pollset_work` on a given pollset are all queued in a linked-list against the `grpc_pollset`. The head of the linked list is called "root worker" 116 117- There are as many neighborhoods as the number of cores. A pollset is put in a neighborhood based on the CPU core of the root worker thread. When picking the next designated poller, we always try to find another worker on the current pollset. If there are no more workers in the current pollset, a `pollset_neighborhood` listed is scanned to pick the next pollset and worker that could be the new designated poller. 118 - NOTE: There is room to tune this implementation. All we really need is good way to maintain a list of `grpc_pollset_workers` with a way to group them per-pollset (needed to implement `grpc_pollset_kick` semantics) and a way randomly select a new designated poller 119 120- See [`begin_worker()`](https://github.com/grpc/grpc/blob/v1.15.1/src/core/lib/iomgr/ev_epoll1_linux.cc#L729) function to see how a designated poller is chosen. Similarly [`end_worker()`](https://github.com/grpc/grpc/blob/v1.15.1/src/core/lib/iomgr/ev_epoll1_linux.cc#L916) function is called by the worker that was just out of `epoll_wait()` and will have to choose a new designated poller) 121 122 123### epollex 124 125 126 127Code at `src/core/lib/iomgr/ev_epollex_posix.cc` 128 129- FDs are added to multiple epollsets with EPOLLEXCLUSIVE flag. This prevents multiple worker threads from waking up from polling whenever the fd is readable/writable 130 131- A few observations: 132 133 - If multiple pollsets are pointing to the same `Pollable`, then the `pollable` MUST be either empty or of type `PO_FD` (i.e single-fd) 134 - A multi-pollable has one-and-only-one incoming link from a pollset 135 - The same FD can be in multiple `Pollable`s (even if one of the `Pollable`s is of type PO_FD) 136 - There cannot be two `Pollable`s of type PO_FD for the same fd 137 138- Why do we need `Pollable` of type PO_FD and PO_EMPTY ? 139 - The main reason is the Sync client API 140 - We create one new completion queue per call. If we didn’t have PO_EMPTY and PO_FD type pollables, then every call on a given channel will effectively have to create a `Pollable` and hence an epollset. This is because every completion queue automatically creates a pollset and the channel fd will have to be put in that pollset. This clearly requires an epollset to put that fd. Creating an epollset per call (even if we delete the epollset once the call is completed) would mean a lot of sys calls to create/delete epoll fds. This is clearly not a good idea. 141 - With these new types of `Pollable`s, all pollsets (corresponding to the new per-call completion queue) will initially point to PO_EMPTY global epollset. Then once the channel fd is added to the pollset, the pollset will point to the `Pollable` of type PO_FD containing just that fd (i.e it will reuse the existing `Pollable`). This way, the epoll fd creation/deletion churn is avoided. 142 143 144### Other polling engine implementations (poll and windows polling engine) 145- **poll** polling engine: gRPC's `poll` polling engine is quite complicated. It uses the `poll()` function to do the polling (and hence it is for platforms like osx where epoll is not available) 146 - The implementation is further complicated by the fact that poll() is level triggered (just keep this in mind in case you wonder why the code at `src/core/lib/iomgr/ev_poll_posix.cc` is written a certain/seemingly complicated way :)) 147 148- **Polling engine on Windows**: Windows polling engine looks nothing like other polling engines 149 - Unlike the grpc polling engines for Unix systems (epollex, epoll1 and poll) Windows endpoint implementation and polling engine implementations are very closely tied together 150 - Windows endpoint read/write API implementations use the Windows IO API which require specifying an [I/O completion port](https://docs.microsoft.com/en-us/windows/desktop/fileio/i-o-completion-ports) 151 - In Windows polling engine’s grpc_pollset_work() implementation, ONE of the threads is chosen to wait on the I/O completion port while other threads wait on a condition variable (much like the turnstile polling in epollex/epoll1) 152 153
