1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357
// Copyright Materialize, Inc. and contributors. All rights reserved.
//
// Use of this software is governed by the Business Source License
// included in the LICENSE file.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0.
//! TopK execution logic.
//!
//! Consult [TopKPlan] documentation for details.
use differential_dataflow::hashable::Hashable;
use differential_dataflow::lattice::Lattice;
use differential_dataflow::operators::arrange::ArrangeBySelf;
use differential_dataflow::operators::reduce::ReduceCore;
use differential_dataflow::operators::Consolidate;
use differential_dataflow::trace::implementations::ord::OrdValSpine;
use differential_dataflow::AsCollection;
use differential_dataflow::Collection;
use timely::dataflow::Scope;
use dataflow_types::plan::top_k::{BasicTopKPlan, MonotonicTop1Plan, MonotonicTopKPlan, TopKPlan};
use repr::{Diff, Row};
use crate::render::context::CollectionBundle;
use crate::render::context::Context;
// The implementation requires integer timestamps to be able to delay feedback for monotonic inputs.
impl<G> Context<G, Row, repr::Timestamp>
where
G: Scope<Timestamp = repr::Timestamp>,
{
pub fn render_topk(
&mut self,
input: CollectionBundle<G, Row, G::Timestamp>,
top_k_plan: TopKPlan,
) -> CollectionBundle<G, Row, G::Timestamp> {
let (ok_input, err_input) = input.as_specific_collection(None);
// We create a new region to compartmentalize the topk logic.
let ok_result = ok_input.scope().region_named("TopK", |inner| {
let ok_input = ok_input.enter(inner);
let ok_result = match top_k_plan {
TopKPlan::MonotonicTop1(MonotonicTop1Plan {
group_key,
order_key,
}) => render_top1_monotonic(ok_input, group_key, order_key),
TopKPlan::MonotonicTopK(MonotonicTopKPlan {
order_key,
group_key,
arity,
limit,
}) => {
// For monotonic inputs, we are able to retract inputs that can no longer be produced
// as outputs. Any inputs beyond `offset + limit` will never again be produced as
// outputs, and can be removed. The simplest form of this is when `offset == 0` and
// these removeable records are those in the input not produced in the output.
// TODO: consider broadening this optimization to `offset > 0` by first filtering
// down to `offset = 0` and `limit = offset + limit`, followed by a finishing act
// of `offset` and `limit`, discarding only the records not produced in the intermediate
// stage.
use differential_dataflow::operators::iterate::Variable;
let delay = std::time::Duration::from_nanos(10_000_000_000);
let retractions =
Variable::new(&mut ok_input.scope(), delay.as_millis() as u64);
let thinned = ok_input.concat(&retractions.negate());
let result = build_topk(thinned, group_key, order_key, 0, limit, arity);
retractions.set(&ok_input.concat(&result.negate()));
result
}
TopKPlan::Basic(BasicTopKPlan {
group_key,
order_key,
offset,
limit,
arity,
}) => build_topk(ok_input, group_key, order_key, offset, limit, arity),
};
// Extract the results from the region.
ok_result.leave_region()
});
return CollectionBundle::from_collections(ok_result, err_input);
/// Constructs a TopK dataflow subgraph.
fn build_topk<G>(
collection: Collection<G, Row, Diff>,
group_key: Vec<usize>,
order_key: Vec<expr::ColumnOrder>,
offset: usize,
limit: Option<usize>,
arity: usize,
) -> Collection<G, Row, Diff>
where
G: Scope,
G::Timestamp: Lattice,
{
let mut datum_vec = repr::DatumVec::new();
let mut collection = collection.map({
move |row| {
let row_hash = row.hashed();
let group_row = {
let datums = datum_vec.borrow_with(&row);
let iterator = group_key.iter().map(|i| datums[*i]);
let total_size = repr::datums_size(iterator.clone());
let mut group_row = Row::with_capacity(total_size);
group_row.extend(iterator);
group_row
};
((group_row, row_hash), row)
}
});
// This sequence of numbers defines the shifts that happen to the 64 bit hash
// of the record, and has the properties that 1. there are not too many of them,
// and 2. each has a modest difference to the next.
//
// These two properties mean that there should be no reductions on groups that
// are substantially larger than `offset + limit` (the largest factor should be
// bounded by two raised to the difference between subsequent numbers);
if let Some(limit) = limit {
for log_modulus in
[60, 56, 52, 48, 44, 40, 36, 32, 28, 24, 20, 16, 12, 8, 4u64].iter()
{
// here we do not apply `offset`, but instead restrict ourself with a limit
// that includes the offset. We cannot apply `offset` until we perform the
// final, complete reduction.
collection = build_topk_stage(
collection,
order_key.clone(),
1u64 << log_modulus,
0,
Some(offset + limit),
arity,
);
}
}
// We do a final step, both to make sure that we complete the reduction, and to correctly
// apply `offset` to the final group, as we have not yet been applying it to the partially
// formed groups.
build_topk_stage(collection, order_key, 1u64, offset, limit, arity)
.map(|((_key, _hash), row)| row)
}
// To provide a robust incremental orderby-limit experience, we want to avoid grouping
// *all* records (or even large groups) and then applying the ordering and limit. Instead,
// a more robust approach forms groups of bounded size (here, 16) and applies the offset
// and limit to each, and then increases the sizes of the groups.
// Builds a "stage", which uses a finer grouping than is required to reduce the volume of
// updates, and to reduce the amount of work on the critical path for updates. The cost is
// a larger number of arrangements when this optimization does nothing beneficial.
fn build_topk_stage<G>(
collection: Collection<G, ((Row, u64), Row), Diff>,
order_key: Vec<expr::ColumnOrder>,
modulus: u64,
offset: usize,
limit: Option<usize>,
arity: usize,
) -> Collection<G, ((Row, u64), Row), Diff>
where
G: Scope,
G::Timestamp: Lattice,
{
use differential_dataflow::operators::Reduce;
let input = collection.map(move |((key, hash), row)| ((key, hash % modulus), row));
// We only want to arrange parts of the input that are not part of the actual output
// such that `input.concat(&negated_output.negate())` yields the correct TopK
let negated_output = input.reduce_named("TopK", {
move |_key, source, target: &mut Vec<(Row, isize)>| {
// Determine if we must actually shrink the result set.
let must_shrink = offset > 0
|| limit
.map(|l| source.iter().map(|(_, d)| *d).sum::<isize>() as usize > l)
.unwrap_or(false);
if must_shrink {
// First go ahead and emit all records
for (row, diff) in source.iter() {
target.push(((*row).clone(), diff.clone()));
}
// local copies that may count down to zero.
let mut offset = offset;
let mut limit = limit;
// The order in which we should produce rows.
let mut indexes = (0..source.len()).collect::<Vec<_>>();
if !order_key.is_empty() {
// We decode the datums once, into a common buffer for efficiency.
// Each row should contain `arity` columns; we should check that.
let mut buffer = Vec::with_capacity(arity * source.len());
for (index, row) in source.iter().enumerate() {
buffer.extend(row.0.iter());
assert_eq!(buffer.len(), arity * (index + 1));
}
let width = buffer.len() / source.len();
//todo: use arrangements or otherwise make the sort more performant?
indexes.sort_by(|left, right| {
let left = &buffer[left * width..][..width];
let right = &buffer[right * width..][..width];
expr::compare_columns(&order_key, left, right, || left.cmp(right))
});
}
// We now need to lay out the data in order of `buffer`, but respecting
// the `offset` and `limit` constraints.
for index in indexes.into_iter() {
let (row, mut diff) = source[index];
if diff > 0 {
// If we are still skipping early records ...
if offset > 0 {
let to_skip = std::cmp::min(offset, diff as usize);
offset -= to_skip;
diff -= to_skip as isize;
}
// We should produce at most `limit` records.
if let Some(limit) = &mut limit {
diff = std::cmp::min(diff, *limit as isize);
*limit -= diff as usize;
}
// Output the indicated number of rows.
if diff > 0 {
// Emit retractions for the elements actually part of
// the set of TopK elements.
target.push((row.clone(), -diff));
}
}
}
}
}
});
negated_output.negate().concat(&input).consolidate()
}
fn render_top1_monotonic<G>(
collection: Collection<G, Row, Diff>,
group_key: Vec<usize>,
order_key: Vec<expr::ColumnOrder>,
) -> Collection<G, Row, Diff>
where
G: Scope,
G::Timestamp: Lattice,
{
// We can place our rows directly into the diff field, and only keep the relevant one
// corresponding to evaluating our aggregate, instead of having to do a hierarchical
// reduction.
use timely::dataflow::operators::Map;
let collection = collection.map({
let mut datum_vec = repr::DatumVec::new();
move |row| {
let group_key = {
let datums = datum_vec.borrow_with(&row);
let iterator = group_key.iter().map(|i| datums[*i]);
let total_size = repr::datums_size(iterator.clone());
let mut group_key = Row::with_capacity(total_size);
group_key.extend(iterator);
group_key
};
(group_key, row)
}
});
// We arrange the inputs ourself to force it into a leaner structure because we know we
// won't care about values.
//
// TODO: Could we use explode here? We'd lose the diff>0 assert and we'd have to impl Mul
// for the monoid, unclear if it's worth it.
let partial: Collection<G, Row, monoids::Top1Monoid> = collection
.consolidate()
.inner
.map(move |((group_key, row), time, diff)| {
assert!(diff > 0);
// NB: Top1 can throw out the diff since we've asserted that it's > 0. A more
// general TopK monoid would have to account for diff.
(
group_key,
time,
monoids::Top1Monoid {
row,
order_key: order_key.clone(),
},
)
})
.as_collection();
let result = partial
.arrange_by_self()
.reduce_abelian::<_, OrdValSpine<_, _, _, _>>("Top1Monotonic", {
move |_key, input, output| {
let accum = &input[0].1;
output.push((accum.row.clone(), 1));
}
});
// TODO(#7331): Here we discard the arranged output.
result.as_collection(|_k, v| v.clone())
}
}
}
/// Monoids for in-place compaction of monotonic streams.
pub mod monoids {
use std::cmp::Ordering;
use differential_dataflow::difference::Semigroup;
use serde::{Deserialize, Serialize};
use expr::ColumnOrder;
use repr::Row;
/// A monoid containing a row and an ordering.
#[derive(Eq, PartialEq, Debug, Clone, Serialize, Deserialize, Hash)]
pub struct Top1Monoid {
pub row: Row,
pub order_key: Vec<ColumnOrder>,
}
impl Ord for Top1Monoid {
fn cmp(&self, other: &Self) -> Ordering {
debug_assert_eq!(self.order_key, other.order_key);
// It might be nice to cache this row decoding like the non-monotonic codepath, but we'd
// have to store the decoded Datums in the same struct as the Row, which gets tricky.
let left: Vec<_> = self.row.unpack();
let right: Vec<_> = other.row.unpack();
expr::compare_columns(&self.order_key, &left, &right, || left.cmp(&right))
}
}
impl PartialOrd for Top1Monoid {
fn partial_cmp(&self, other: &Self) -> Option<Ordering> {
Some(self.cmp(other))
}
}
impl Semigroup for Top1Monoid {
fn plus_equals(&mut self, rhs: &Self) {
// It's unclear what the semantics are of a TopK without an ordering, but match the
// non-monotonic impl's behavior of not doing any decoding work.
if self.order_key.is_empty() {
return;
}
let cmp = (&*self).cmp(rhs);
// NB: Reminder that TopK returns the _minimum_ K items.
if cmp == Ordering::Greater {
self.clone_from(&rhs);
}
}
fn is_zero(&self) -> bool {
false
}
}
}