tustvold commented on code in PR #4522: URL: https://github.com/apache/arrow-datafusion/pull/4522#discussion_r1051388943
########## datafusion/core/src/execution/memory_pool/pool.rs: ########## @@ -0,0 +1,280 @@ +// Licensed to the Apache Software Foundation (ASF) under one +// or more contributor license agreements. See the NOTICE file +// distributed with this work for additional information +// regarding copyright ownership. The ASF licenses this file +// to you 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. + +use crate::execution::memory_pool::{AllocationOptions, MemoryPool, TrackedAllocation}; +use datafusion_common::{DataFusionError, Result}; +use parking_lot::Mutex; +use std::sync::atomic::{AtomicUsize, Ordering}; + +/// A [`MemoryPool`] that enforces no limit +#[derive(Debug, Default)] +pub struct UnboundedMemoryPool { + used: AtomicUsize, +} + +impl MemoryPool for UnboundedMemoryPool { + fn grow(&self, _allocation: &TrackedAllocation, additional: usize) { + self.used.fetch_add(additional, Ordering::Relaxed); + } + + fn shrink(&self, _allocation: &TrackedAllocation, shrink: usize) { + self.used.fetch_sub(shrink, Ordering::Relaxed); + } + + fn try_grow(&self, allocation: &TrackedAllocation, additional: usize) -> Result<()> { + self.grow(allocation, additional); + Ok(()) + } + + fn allocated(&self) -> usize { + self.used.load(Ordering::Relaxed) + } +} + +/// A [`MemoryPool`] that implements a greedy first-come first-serve limit +#[derive(Debug)] +pub struct GreedyMemoryPool { + pool_size: usize, + used: AtomicUsize, +} + +impl GreedyMemoryPool { + /// Allocate up to `limit` bytes + pub fn new(pool_size: usize) -> Self { + Self { + pool_size, + used: AtomicUsize::new(0), + } + } +} + +impl MemoryPool for GreedyMemoryPool { + fn grow(&self, _allocation: &TrackedAllocation, additional: usize) { + self.used.fetch_add(additional, Ordering::Relaxed); + } + + fn shrink(&self, _allocation: &TrackedAllocation, shrink: usize) { + self.used.fetch_sub(shrink, Ordering::Relaxed); + } + + fn try_grow(&self, allocation: &TrackedAllocation, additional: usize) -> Result<()> { + self.used + .fetch_update(Ordering::Relaxed, Ordering::Relaxed, |used| { + let new_used = used + additional; + (new_used <= self.pool_size).then_some(new_used) + }) + .map_err(|used| { + insufficient_capacity_err(allocation, additional, self.pool_size - used) + })?; + Ok(()) + } + + fn allocated(&self) -> usize { + self.used.load(Ordering::Relaxed) + } +} + +/// A [`MemoryPool`] that prevents spillable allocations from using more than +/// an even fraction of the available memory sans any unspillable allocations +/// (i.e. `(pool_size - unspillable_memory) / num_spillable_allocations`) +/// +/// ┌───────────────────────z──────────────────────z───────────────┐ +/// │ z z │ +/// │ z z │ +/// │ Spillable z Unspillable z Free │ +/// │ Memory z Memory z Memory │ +/// │ z z │ +/// │ z z │ +/// └───────────────────────z──────────────────────z───────────────┘ +/// +/// Unspillable memory is allocated in a first-come, first-serve fashion +#[derive(Debug)] +pub struct FairSpillPool { + /// The total memory limit + pool_size: usize, + + state: Mutex<FairSpillPoolState>, +} + +#[derive(Debug)] +struct FairSpillPoolState { + /// The number of allocations that can spill + num_spill: usize, + + /// The total amount of memory allocated that can be spilled + spillable: usize, + + /// The total amount of memory allocated by consumers that cannot spill + unspillable: usize, +} + +impl FairSpillPool { + /// Allocate up to `limit` bytes + pub fn new(pool_size: usize) -> Self { + Self { + pool_size, + state: Mutex::new(FairSpillPoolState { + num_spill: 0, + spillable: 0, + unspillable: 0, + }), + } + } +} + +impl MemoryPool for FairSpillPool { + fn allocate(&self, options: &AllocationOptions) { Review Comment: > I believe it will be fine because each ExternalSorter creates the potentially spilling allocation allocation upon construction, though I am still not 100% sure. Now that ExecutionPlan is sync I don't think this is an issue, the streams are all created prior to execution starting. Yes theoretically you could lazily construct the `TrackedAllocation`, I can't think of why you would structure your code in this way. Regardless this issue existed before the changes in this PR, it is just easier to see > That way the intent would be clearer that all ExternalSort instances register themselves with the MemoryManager on creation, and part of registering would increase the num_spill count. Perhaps I'm being stupid but I don't see a tangible difference between `TrackedAllocation::new(pool, options)` and `pool.register(options)`, you're just reordering the parameters? More specifically in order for `register` to be implemented, you still need a public `TrackedAllocation::new` so you're just adding an indirection? -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
