Chiebot-Mirror/grpc - grpc - Gitea: Git with a cup of tea

Commit Graph

Author	SHA1	Message	Date
Craig Tiller	7e089a22d7	[party] Make it faster (#37132 ) Notes: * Adds a single participant `AddParticipant` variant for this common case (per #37056 which I'm abandoning) * Folds the `PartySyncUsingAtomics` class back into `Party`, removes the `PartySyncUsingMutex` class * Leverages this integration to find places where we're doing repeated CAS operations and folds them into single operations (for example an unlock/unref pair can be folded into a single CAS) * Also lowers some `CHECK` statements into `DCHECK` - which I think is appropriate given the performance sensitivity of this code * Adds code to deal with overflowing the number of participants added to a party -- for now we do a busy add by queuing to event engine and retrying -- this has the advantage of not adding cost to the normal path, but has the slightly worrying disadvantage of effectively being a busy poll. My expectation is that this will be ok in general (the condition clears very quickly), but if not we'll modify this to be a linked list of pending actions and take a hit on the fast path. * Simplifies `PartyIsOver` (per #37113 which I'm abandoning) * Keeps a per-object wakeup cache (`wakeup_mask_`) that is protected by the lock bit -- this allows waking up a participant during polling without resorting to an extra atomic operation - significantly speeding that wakeup path (17ns --> 6ns) Before: ``` ---------------------------------------------------------------------- Benchmark Time CPU Iterations ---------------------------------------------------------------------- BM_PartyCreate 142 ns 142 ns 44952269 BM_PartyCreate 73.8 ns 73.8 ns 44952269 BM_PartyCreate 72.6 ns 72.6 ns 44952269 BM_PartyCreate 72.5 ns 72.5 ns 44952269 BM_PartyCreate 72.4 ns 72.4 ns 44952269 BM_PartyCreate 72.5 ns 72.5 ns 44952269 BM_PartyCreate 72.5 ns 72.5 ns 44952269 BM_PartyCreate 72.6 ns 72.6 ns 44952269 BM_PartyCreate 72.2 ns 72.2 ns 44952269 BM_PartyCreate 72.5 ns 72.5 ns 44952269 BM_PartyCreate_mean 79.5 ns 79.5 ns 10 BM_PartyCreate_median 72.5 ns 72.5 ns 10 BM_PartyCreate_stddev 21.8 ns 21.8 ns 10 BM_PartyCreate_cv 27.46 % 27.46 % 10 BM_AddParticipant 35.3 ns 35.3 ns 197041251 BM_AddParticipant 35.3 ns 35.3 ns 197041251 BM_AddParticipant 35.1 ns 35.1 ns 197041251 BM_AddParticipant 35.4 ns 35.4 ns 197041251 BM_AddParticipant 35.3 ns 35.3 ns 197041251 BM_AddParticipant 35.2 ns 35.2 ns 197041251 BM_AddParticipant 35.9 ns 35.9 ns 197041251 BM_AddParticipant 36.0 ns 36.0 ns 197041251 BM_AddParticipant 35.8 ns 35.8 ns 197041251 BM_AddParticipant 36.0 ns 36.0 ns 197041251 BM_AddParticipant_mean 35.5 ns 35.5 ns 10 BM_AddParticipant_median 35.4 ns 35.4 ns 10 BM_AddParticipant_stddev 0.352 ns 0.352 ns 10 BM_AddParticipant_cv 0.99 % 0.99 % 10 BM_WakeupParticipant 17.1 ns 17.1 ns 406116840 BM_WakeupParticipant 16.9 ns 16.9 ns 406116840 BM_WakeupParticipant 16.8 ns 16.8 ns 406116840 BM_WakeupParticipant 16.8 ns 16.8 ns 406116840 BM_WakeupParticipant 16.9 ns 16.9 ns 406116840 BM_WakeupParticipant 16.9 ns 16.9 ns 406116840 BM_WakeupParticipant 17.0 ns 17.0 ns 406116840 BM_WakeupParticipant 17.0 ns 17.0 ns 406116840 BM_WakeupParticipant 16.9 ns 16.9 ns 406116840 BM_WakeupParticipant 17.0 ns 17.0 ns 406116840 BM_WakeupParticipant_mean 16.9 ns 16.9 ns 10 BM_WakeupParticipant_median 16.9 ns 16.9 ns 10 BM_WakeupParticipant_stddev 0.087 ns 0.087 ns 10 BM_WakeupParticipant_cv 0.51 % 0.51 % 10 ``` After: ``` ---------------------------------------------------------------------- Benchmark Time CPU Iterations ---------------------------------------------------------------------- BM_PartyCreate 115 ns 115 ns 29602192 BM_PartyCreate 56.5 ns 56.5 ns 29602192 BM_PartyCreate 55.3 ns 55.3 ns 29602192 BM_PartyCreate 55.9 ns 55.9 ns 29602192 BM_PartyCreate 55.1 ns 55.1 ns 29602192 BM_PartyCreate 55.2 ns 55.2 ns 29602192 BM_PartyCreate 55.2 ns 55.2 ns 29602192 BM_PartyCreate 56.2 ns 56.2 ns 29602192 BM_PartyCreate 54.7 ns 54.7 ns 29602192 BM_PartyCreate 55.8 ns 55.8 ns 29602192 BM_PartyCreate_mean 61.5 ns 61.5 ns 10 BM_PartyCreate_median 55.5 ns 55.5 ns 10 BM_PartyCreate_stddev 18.9 ns 18.9 ns 10 BM_PartyCreate_cv 30.68 % 30.68 % 10 BM_AddParticipant 26.9 ns 26.9 ns 155407231 BM_AddParticipant 26.5 ns 26.5 ns 155407231 BM_AddParticipant 24.8 ns 24.8 ns 155407231 BM_AddParticipant 24.9 ns 24.9 ns 155407231 BM_AddParticipant 24.8 ns 24.8 ns 155407231 BM_AddParticipant 25.3 ns 25.3 ns 155407231 BM_AddParticipant 25.8 ns 25.8 ns 155407231 BM_AddParticipant 25.3 ns 25.3 ns 155407231 BM_AddParticipant 30.8 ns 30.8 ns 155407231 BM_AddParticipant 27.7 ns 27.7 ns 155407231 BM_AddParticipant_mean 26.3 ns 26.3 ns 10 BM_AddParticipant_median 25.6 ns 25.6 ns 10 BM_AddParticipant_stddev 1.87 ns 1.87 ns 10 BM_AddParticipant_cv 7.11 % 7.10 % 10 BM_WakeupParticipant 6.75 ns 6.75 ns 623459241 BM_WakeupParticipant 6.77 ns 6.77 ns 623459241 BM_WakeupParticipant 6.74 ns 6.74 ns 623459241 BM_WakeupParticipant 6.73 ns 6.73 ns 623459241 BM_WakeupParticipant 6.74 ns 6.74 ns 623459241 BM_WakeupParticipant 6.70 ns 6.70 ns 623459241 BM_WakeupParticipant 6.70 ns 6.69 ns 623459241 BM_WakeupParticipant 6.79 ns 6.79 ns 623459241 BM_WakeupParticipant 6.76 ns 6.76 ns 623459241 BM_WakeupParticipant 6.78 ns 6.78 ns 623459241 BM_WakeupParticipant_mean 6.75 ns 6.75 ns 10 BM_WakeupParticipant_median 6.75 ns 6.75 ns 10 BM_WakeupParticipant_stddev 0.031 ns 0.031 ns 10 BM_WakeupParticipant_cv 0.46 % 0.46 % 10 ``` Closes #37132 COPYBARA_INTEGRATE_REVIEW=https://github.com/grpc/grpc/pull/37132 from ctiller:nineteen-ninety-nine `336c87bdd6` PiperOrigin-RevId: 656437265	5 months ago
Craig Tiller	d5fa1814b6	[party] Eliminate the `AfterCurrentPoll` mechanism (#37036 ) We're not using this right now and so let's eliminate it until we've got a solid use-case. `bm_party` results -- before: ``` --------------------------------------------------------------- Benchmark Time CPU Iterations --------------------------------------------------------------- BM_PartyCreate 39.4 ns 39.4 ns 106472775 BM_AddParticipant 44.2 ns 44.2 ns 94802899 BM_WakeupParticipant 17.4 ns 17.4 ns 241990117 ``` and after: ``` --------------------------------------------------------------- Benchmark Time CPU Iterations --------------------------------------------------------------- BM_PartyCreate 37.6 ns 37.6 ns 111332125 BM_AddParticipant 40.1 ns 40.1 ns 104740937 BM_WakeupParticipant 17.3 ns 17.3 ns 242484270 ``` Closes #37036 COPYBARA_INTEGRATE_REVIEW=https://github.com/grpc/grpc/pull/37036 from ctiller:no-after `42f396900f` PiperOrigin-RevId: 646994419	6 months ago
Craig Tiller	678ee0a45d	[party] Make party a concrete type (#37015 ) Move event engine context to be an arena based context, and have party host arena directly. Now `Party` can directly inject all context necessary without needing to call out to the derived type; this makes `Party` instantiable by itself (no need to inherit from it - though we'll still want to for `CallSpine` to simplify some arena lifetime management), meaning we can start to consider eliminating the base class `Activity` and merging these types. Closes #37015 COPYBARA_INTEGRATE_REVIEW=https://github.com/grpc/grpc/pull/37015 from ctiller:chaotic-party `7f44b37b5d` PiperOrigin-RevId: 646963003	6 months ago
Tanvi Jagtap	ed5893120e	[grpc][Gpr_To_Absl_Logging] Migrating from gpr to absl logging - gpr_log (#36608 ) [grpc][Gpr_To_Absl_Logging] Migrating from gpr to absl logging - gpr_log In this CL we are migrating from gRPCs own gpr logging mechanism to absl logging mechanism. The intention is to deprecate gpr_log in the future. We have the following mapping 1. gpr_log(GPR_INFO,...) -> LOG(INFO) 2. gpr_log(GPR_ERROR,...) -> LOG(ERROR) 3. gpr_log(GPR_DEBUG,...) -> VLOG(2) Reviewers need to check : 1. If the above mapping is correct. 2. The content of the log is as before. gpr_log format strings did not use string_view or std::string . absl LOG accepts these. So there will be some elimination of string_view and std::string related conversions. This is expected. Closes #36608 COPYBARA_INTEGRATE_REVIEW=https://github.com/grpc/grpc/pull/36608 from tanvi-jagtap:regex_test_core_misc `e6b9481dac` PiperOrigin-RevId: 634246311	7 months ago
Craig Tiller	e0cade4daa	[party] Decouple `Party` from `Arena` (#36229 ) Following up to #33961 `Party` no longer needs to refer to `Arena`, and decoupling gives us a few more degrees of freedom in the design of a final `CallSpine`. Also flesh out `LogStateChange` usage so that all state transitions are traced when that tracer is enabled. Closes #36229 COPYBARA_INTEGRATE_REVIEW=https://github.com/grpc/grpc/pull/36229 from ctiller:arenaless-party `51ae8eb898` PiperOrigin-RevId: 621525912	8 months ago
Craig Tiller	15d64cf9c9	[party] Auto-batch cross party wakeups (#36048 ) If party1 wakes party2 we'd prefer that party1 finishes all its activations before it wakes party2: intuition: party1 might wake party2 for other things too, and being able to coalesce them all into one wakeup significantly helps performance. Closes #36048 COPYBARA_INTEGRATE_REVIEW=https://github.com/grpc/grpc/pull/36048 from ctiller:wakey `f6d4416685` PiperOrigin-RevId: 616970833	9 months ago
Craig Tiller	67d6b8ea74	[promise] Context improvements (#35592 ) A few improvements to the promise context system (more coming) Allow subclassed contexts: If we have multiple different kinds of a base context, allow `GetContext<Derived>()` to mean `down_cast<Derived>(GetContext<Base>())` everywhere for brevity. Allow custom context lookup: For a base context type, allow customization of how that context is looked up. These two together allow: 1. normalization of activity lookup and context lookup to the same syntax (so we can write `GetContext<Activity>()` everywhere now 2. Party & Activity to share a context, so that anywhere we need to do a party specific operation we can write `GetContext<Party>()->...` and safely know that it's the current activity and* it's a party. Closes #35592 COPYBARA_INTEGRATE_REVIEW=https://github.com/grpc/grpc/pull/35592 from ctiller:contextual-types `37ef948a36` PiperOrigin-RevId: 599651708	11 months ago
Craig Tiller	f82f8966f0	[promise] Add a facility to delay promise execution until after the current poll (#35413 ) Mirrors what we had with combiner, but allows it to occur at arbitrary points. We'll use this in chaotic-good to: 1. combine fragments into a single frame 2. combine writes from different calls into a single syscall Closes #35413 COPYBARA_INTEGRATE_REVIEW=https://github.com/grpc/grpc/pull/35413 from ctiller:group-on `9f20f34523` PiperOrigin-RevId: 596004767	11 months ago
Craig Tiller	f11c010e51	[promises] Add awaitable spawn to Party (#34744 ) Add a variant of `Spawn` that returns a promise that can be awaited by another activity. This allows us to simply implement complex cross-activity synchronization. (necessary building block for #34740) Also adds an inter-activity latch as a building block to test this work. Closes #34744 COPYBARA_INTEGRATE_REVIEW=https://github.com/grpc/grpc/pull/34744 from ctiller:ninteen-ninety-nine `19074b255f` PiperOrigin-RevId: 582450643	1 year ago
Craig Tiller	4d0611e4a9	[promises] Async wakeup for activities/parties (#33215 ) Parties prefer to wakeup inline, however there are some mechanisms that want an out-of-line wakeup (say due to a previously held mutex that may be re-taken). To help those cases permit a guaranteed asynchronous wakeup. (needed now for resolver wakeups on the client call path)	2 years ago
Craig Tiller	dcef4bb981	[party] Disable mutex test on mac (#33143 ) This has been flaky, but we don't need this scenario tested right now (especially on Mac), so disable that part of the testing.	2 years ago
Craig Tiller	8d2f70d53c	Reland "[promises] Convert call to a party" (#32651 )" (#32653 ) <!-- If you know who should review your pull request, please assign it to that person, otherwise the pull request would get assigned randomly. If your pull request is for a specific language, please add the appropriate lang label. --> --------- Co-authored-by: ctiller <ctiller@users.noreply.github.com>	2 years ago
Yash Tibrewal	29ce6463d1	Revert "[promises] Convert call to a party" (#32651 ) Reverts grpc/grpc#32359	2 years ago
Craig Tiller	a9873e8357	[promises] Convert call to a party (#32359 ) <!-- If you know who should review your pull request, please assign it to that person, otherwise the pull request would get assigned randomly. If your pull request is for a specific language, please add the appropriate lang label. --> --------- Co-authored-by: ctiller <ctiller@users.noreply.github.com>	2 years ago
Craig Tiller	0d7b34451a	[promises] Make Poll<T> its own type, not a variant<> (#32515 ) Avoids some compilation problems on older MSVC's, opens the door for some future optimizations. --------- Co-authored-by: ctiller <ctiller@users.noreply.github.com>	2 years ago
Craig Tiller	0ecc18ef0f	[promises] Party: an activity with many participant promises (#32308 ) <!-- If you know who should review your pull request, please assign it to that person, otherwise the pull request would get assigned randomly. If your pull request is for a specific language, please add the appropriate lang label. --> --------- Co-authored-by: ctiller <ctiller@users.noreply.github.com>	2 years ago

16 Commits (60fbdef6767a71e6a29576488bddc24c798ea94e)