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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923
// 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.
//! Types and methods related to initializing, updating, and removing read policies
//! on collections.
//!
//! This module contains the API for read holds on collections. A "read hold" prevents
//! the controller from compacting the associated collections, and ensures that they
//! remain "readable" at a specific time, as long as the hold is held.
//! Allow usage of `std::collections::HashMap`.
//! The code in this module deals with `Antichain`-keyed maps. `Antichain` does not implement
//! `Ord`, so we cannot use `BTreeMap`s. We need to iterate through the maps, so we cannot use the
//! `mz_ore` wrapper either.
#![allow(clippy::disallowed_types)]
use std::collections::{btree_map, BTreeMap, BTreeSet, HashMap};
use std::fmt::Debug;
use std::hash::Hash;
use std::ops::Deref;
use differential_dataflow::lattice::Lattice;
use itertools::Itertools;
use mz_adapter_types::compaction::{CompactionWindow, ReadCapability};
use mz_compute_types::ComputeInstanceId;
use mz_ore::instrument;
use mz_repr::{GlobalId, Timestamp};
use mz_sql::session::metadata::SessionMetadata;
use mz_storage_types::read_policy::ReadPolicy;
use serde::Serialize;
use timely::progress::frontier::MutableAntichain;
use timely::progress::Antichain;
use timely::progress::Timestamp as TimelyTimestamp;
use crate::coord::id_bundle::CollectionIdBundle;
use crate::coord::timeline::{TimelineContext, TimelineState};
use crate::coord::Coordinator;
use crate::session::Session;
use crate::util::ResultExt;
/// For each timeline, we hold one [TimelineReadHolds] as the root read holds
/// for that timeline. Even if there are no other read holds ([ReadHolds] and/or
/// [ReadHoldsInner]), it acts as a backstop that makes sure that collections
/// remain readable at the read timestamp (according to the
/// timestamp oracle) of that timeline.
///
/// When creating a collection in a timeline, it is added to the one
/// [TimelineReadHolds] of that timeline and when a collection is dropped it is
/// removed.
///
/// A [TimelineReadHolds] is never released, it is only dropped when the
/// corresponding timeline is dropped, which only happens when all collections
/// in it have been dropped. We only add collections, remove collections, and
/// downgrade (update, yes!) the read holds.
#[derive(Debug, Serialize)]
pub struct TimelineReadHolds<T> {
pub holds: HashMap<Antichain<T>, CollectionIdBundle>,
}
impl<T: Eq + Hash + Ord> TimelineReadHolds<T> {
/// Return empty `ReadHolds`.
pub fn new() -> Self {
TimelineReadHolds {
holds: HashMap::new(),
}
}
/// Returns whether the [TimelineReadHolds] is empty.
pub fn is_empty(&self) -> bool {
self.holds.is_empty()
}
/// Returns an iterator over all times at which a read hold exists.
pub fn times(&self) -> impl Iterator<Item = &Antichain<T>> {
self.holds.keys()
}
/// Return a `CollectionIdBundle` containing all the IDs in the
/// [TimelineReadHolds].
pub fn id_bundle(&self) -> CollectionIdBundle {
self.holds
.values()
.fold(CollectionIdBundle::default(), |mut accum, id_bundle| {
accum.extend(id_bundle);
accum
})
}
/// Returns an iterator over all storage ids and the time at which their read hold exists.
#[allow(unused)]
pub fn storage_ids(&self) -> impl Iterator<Item = (&Antichain<T>, &GlobalId)> {
self.holds
.iter()
.flat_map(|(time, id_bundle)| std::iter::repeat(time).zip(id_bundle.storage_ids.iter()))
}
/// Returns an iterator over all compute ids by compute instance and the time at which their
/// read hold exists.
pub fn compute_ids(
&self,
) -> impl Iterator<
Item = (
&ComputeInstanceId,
impl Iterator<Item = (&Antichain<T>, &GlobalId)>,
),
> {
let compute_instances: BTreeSet<_> = self
.holds
.iter()
.flat_map(|(_, id_bundle)| id_bundle.compute_ids.keys())
.collect();
compute_instances.into_iter().map(|compute_instance| {
let inner_iter = self
.holds
.iter()
.filter_map(|(time, id_bundle)| {
id_bundle
.compute_ids
.get(compute_instance)
.map(|ids| std::iter::repeat(time).zip(ids.iter()))
})
.flatten();
(compute_instance, inner_iter)
})
}
/// Extends a [TimelineReadHolds] with the contents of another
/// [TimelineReadHolds].
///
/// Asserts that the newly added read holds don't coincide with any of the existing read holds in self.
pub fn extend_with_new(&mut self, mut other: TimelineReadHolds<T>) {
for (time, other_id_bundle) in other.holds.drain() {
let self_id_bundle = self.holds.entry(time).or_default();
assert!(
self_id_bundle.intersection(&other_id_bundle).is_empty(),
"extend_with_new encountered duplicate read holds",
);
self_id_bundle.extend(&other_id_bundle);
}
}
/// If the read hold contains a storage ID equal to `id`, removes it from the read hold and
/// drops it.
pub fn remove_storage_id(&mut self, id: &GlobalId) {
for (_, id_bundle) in &mut self.holds {
id_bundle.storage_ids.remove(id);
}
self.holds.retain(|_, id_bundle| !id_bundle.is_empty());
}
/// If the read hold contains a compute ID equal to `id` in `compute_instance`, removes it from
/// the read hold and drops it.
pub fn remove_compute_id(&mut self, compute_instance: &ComputeInstanceId, id: &GlobalId) {
for (_, id_bundle) in &mut self.holds {
if let Some(compute_ids) = id_bundle.compute_ids.get_mut(compute_instance) {
compute_ids.remove(id);
if compute_ids.is_empty() {
id_bundle.compute_ids.remove(compute_instance);
}
}
}
self.holds.retain(|_, id_bundle| !id_bundle.is_empty());
}
}
/// [ReadHolds] are used for short-lived read holds. For example, when
/// processing peeks or rendering dataflows. These are never downgraded but they
/// _are_ released automatically when being dropped.
pub struct ReadHolds<T: Eq + Hash + Ord> {
pub inner: ReadHoldsInner<T>,
dropped_read_holds_tx: tokio::sync::mpsc::UnboundedSender<ReadHoldsInner<T>>,
}
impl<T: Eq + Hash + Ord> ReadHolds<T> {
/// Return empty `ReadHolds`.
pub fn new(
read_holds: ReadHoldsInner<T>,
dropped_read_holds_tx: tokio::sync::mpsc::UnboundedSender<ReadHoldsInner<T>>,
) -> Self {
ReadHolds {
inner: read_holds,
dropped_read_holds_tx,
}
}
}
impl<T: TimelyTimestamp + Lattice> ReadHolds<T> {
pub fn merge(&mut self, mut other: Self) {
// Now, when other is dropped we don't release the holds anymore.
// Instead we take ownership and move them to ourselves.
let other_inner = std::mem::take(&mut other.inner);
self.inner.merge(other_inner)
}
}
impl<T: Eq + Hash + Ord> Deref for ReadHolds<T> {
type Target = ReadHoldsInner<T>;
fn deref(&self) -> &ReadHoldsInner<T> {
&self.inner
}
}
impl<T: Debug + Eq + Hash + Ord> Debug for ReadHolds<T> {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
f.debug_struct("ReadHolds")
.field("read_holds", &self.inner)
.finish_non_exhaustive()
}
}
impl<T: Eq + Hash + Ord> Drop for ReadHolds<T> {
fn drop(&mut self) {
let inner_holds = std::mem::take(&mut self.inner);
tracing::debug!(
"dropping ReadHolds on storage {:?} and compute {:?} ",
inner_holds.storage_holds.keys(),
inner_holds.compute_holds.keys()
);
let res = self.dropped_read_holds_tx.send(inner_holds);
if let Err(e) = res {
tracing::warn!("error when trying to drop ReadHold: {:?}", e)
}
}
}
/// Inner state of [ReadHolds]. We have this separate so that we can send the
/// inner state along a channel, for releasing when dropped.
#[derive(Debug)]
pub struct ReadHoldsInner<T> {
pub storage_holds: HashMap<GlobalId, MutableAntichain<T>>,
pub compute_holds: HashMap<(ComputeInstanceId, GlobalId), MutableAntichain<T>>,
}
impl<T: Eq + Hash + Ord> ReadHoldsInner<T> {
/// Return empty `ReadHolds`.
pub fn new() -> Self {
ReadHoldsInner {
storage_holds: HashMap::new(),
compute_holds: HashMap::new(),
}
}
/// Return a `CollectionIdBundle` containing all the IDs in the
/// [ReadHoldsInner].
pub fn id_bundle(&self) -> CollectionIdBundle {
let mut res = CollectionIdBundle::default();
for id in self.storage_holds.keys() {
res.storage_ids.insert(*id);
}
for (instance_id, id) in self.compute_holds.keys() {
res.compute_ids.entry(*instance_id).or_default().insert(*id);
}
res
}
}
impl<T: TimelyTimestamp + Lattice> ReadHoldsInner<T> {
pub fn least_valid_read(&self) -> Antichain<T> {
let mut since = Antichain::from_elem(T::minimum());
for (_id, hold) in self.storage_holds.iter() {
since.join_assign(&hold.frontier().to_owned());
}
for (_id, hold) in self.compute_holds.iter() {
since.join_assign(&hold.frontier().to_owned());
}
since
}
/// Returns the frontier at which this [ReadHoldsInner] is holding back the
/// since of the collection identified by `id`. This does not mean that the
/// overall since of the collection is what we report here. Only that it is
/// _at least_ held back to the reported frontier by this read hold.
///
/// This method is not meant to be fast, use wisely!
pub fn since(&self, desired_id: &GlobalId) -> Antichain<T> {
let mut since = Antichain::new();
if let Some(hold) = self.storage_holds.get(desired_id) {
since.extend(hold.frontier().to_owned());
}
for ((_instance, id), hold) in self.compute_holds.iter() {
if id != desired_id {
continue;
}
since.extend(hold.frontier().to_owned());
}
since
}
pub fn merge(&mut self, other: Self) {
for (id, mut other_hold) in other.storage_holds {
let hold = self.storage_holds.entry(id).or_default();
hold.update_iter(other_hold.updates().cloned());
}
for (id, mut other_hold) in other.compute_holds {
let hold = self.compute_holds.entry(id).or_default();
hold.update_iter(other_hold.updates().cloned());
}
}
}
impl<T: Eq + Hash + Ord> Default for ReadHoldsInner<T> {
fn default() -> Self {
ReadHoldsInner::new()
}
}
impl crate::coord::Coordinator {
/// Initialize the storage read policies.
///
/// This should be called only after a storage collection is created, and
/// ideally very soon afterwards. The collection is otherwise initialized
/// with a read policy that allows no compaction.
pub(crate) async fn initialize_storage_read_policies(
&mut self,
ids: Vec<GlobalId>,
compaction_window: CompactionWindow,
) {
self.initialize_read_policies(
&CollectionIdBundle {
storage_ids: ids.into_iter().collect(),
compute_ids: BTreeMap::new(),
},
compaction_window,
)
.await;
}
/// Initialize the compute read policies.
///
/// This should be called only after a compute collection is created, and
/// ideally very soon afterwards. The collection is otherwise initialized
/// with a read policy that allows no compaction.
pub(crate) async fn initialize_compute_read_policies(
&mut self,
ids: Vec<GlobalId>,
instance: ComputeInstanceId,
compaction_window: CompactionWindow,
) {
let mut compute_ids: BTreeMap<_, BTreeSet<_>> = BTreeMap::new();
compute_ids.insert(instance, ids.into_iter().collect());
self.initialize_read_policies(
&CollectionIdBundle {
storage_ids: BTreeSet::new(),
compute_ids,
},
compaction_window,
)
.await;
}
/// Initialize the storage and compute read policies.
///
/// This should be called only after a collection is created, and
/// ideally very soon afterwards. The collection is otherwise initialized
/// with a read policy that allows no compaction.
#[instrument(name = "coord::initialize_read_policies")]
pub(crate) async fn initialize_read_policies(
&mut self,
id_bundle: &CollectionIdBundle,
compaction_window: CompactionWindow,
) {
// Creates a `ReadHolds` struct that contains a read hold for each id in
// `id_bundle`. The time of each read holds is at `time`, if possible
// otherwise it is at the lowest possible time, meaning the implied
// capability of the collection.
//
// This does not apply the read holds in STORAGE or COMPUTE. The code
// below applies those, after ensuring that read capabilities exist.
let initialize_read_holds = |coord: &mut Coordinator,
time: mz_repr::Timestamp,
id_bundle: &CollectionIdBundle|
-> TimelineReadHolds<mz_repr::Timestamp> {
let mut read_holds = TimelineReadHolds::new();
let time = Antichain::from_elem(time);
for id in id_bundle.storage_ids.iter() {
// TODO(aljoscha): This is a bit iffy, because we're using the
// since without having a read hold in place. In this case it's
// fine because we are the one controlling the read policy and
// the storage controller currently cannot advance its frontiers
// concurrently.
//
// This will be fixed properly in a future commit that makes the
// storage controller concurrent.
let (read_frontier, _upper) = coord
.controller
.storage
.collection_frontiers(*id)
.expect("collection does not exist");
let time = time.join(&read_frontier);
read_holds
.holds
.entry(time)
.or_default()
.storage_ids
.insert(*id);
}
for (compute_instance, compute_ids) in id_bundle.compute_ids.iter() {
let compute = coord.controller.active_compute();
for id in compute_ids.iter() {
let collection = compute
.collection(*compute_instance, *id)
.expect("collection does not exist");
let read_frontier = collection.read_capability().clone();
let time = time.join(&read_frontier);
read_holds
.holds
.entry(time)
.or_default()
.compute_ids
.entry(*compute_instance)
.or_default()
.insert(*id);
}
}
read_holds
};
let mut compute_policy_updates: BTreeMap<ComputeInstanceId, Vec<_>> = BTreeMap::new();
let mut storage_policy_updates = Vec::new();
let mut id_bundles: HashMap<_, CollectionIdBundle> = HashMap::new();
// Update the Coordinator's timeline read hold state and organize all id bundles by time.
for (timeline_context, id_bundle) in self.partition_ids_by_timeline_context(id_bundle) {
match timeline_context {
TimelineContext::TimelineDependent(timeline) => {
let TimelineState { oracle, .. } = self.ensure_timeline_state(&timeline).await;
let read_ts = oracle.read_ts().await;
let new_read_holds = initialize_read_holds(self, read_ts, &id_bundle);
let TimelineState { read_holds, .. } =
self.ensure_timeline_state(&timeline).await;
for (time, id_bundle) in &new_read_holds.holds {
id_bundles
.entry(Some(time.clone()))
.or_default()
.extend(id_bundle);
}
read_holds.extend_with_new(new_read_holds);
}
TimelineContext::TimestampIndependent | TimelineContext::TimestampDependent => {
id_bundles.entry(None).or_default().extend(&id_bundle);
}
}
}
// Create read capabilities for all objects.
for (time, id_bundle) in id_bundles {
for (compute_instance, compute_ids) in id_bundle.compute_ids {
for id in compute_ids {
let read_capability = self.ensure_compute_capability(
&compute_instance,
&id,
Some(compaction_window.clone()),
);
if let Some(time) = &time {
read_capability
.holds
.update_iter(time.iter().map(|t| (*t, 1)));
}
compute_policy_updates
.entry(compute_instance)
.or_default()
.push((id, self.compute_read_capabilities[&id].policy()));
}
}
for id in id_bundle.storage_ids {
let read_capability =
self.ensure_storage_capability(&id, Some(compaction_window.clone()));
if let Some(time) = &time {
read_capability
.holds
.update_iter(time.iter().map(|t| (*t, 1)));
}
storage_policy_updates.push((id, self.storage_read_capabilities[&id].policy()));
}
}
// Apply read capabilities.
for (compute_instance, compute_policy_updates) in compute_policy_updates {
self.controller
.active_compute()
.set_read_policy(compute_instance, compute_policy_updates)
.unwrap_or_terminate("cannot fail to set read policy");
}
self.controller
.storage
.set_read_policy(storage_policy_updates);
}
/// Attempt to update the timestamp of the read holds on the indicated collections from the
/// indicated times within `read_holds` to `new_time`.
pub(super) fn update_timeline_read_holds(
&mut self,
read_holds: &mut TimelineReadHolds<mz_repr::Timestamp>,
new_time: mz_repr::Timestamp,
) {
// After this, read_holds.holds is initialized to an empty HashMap.
let old_holds = std::mem::take(&mut read_holds.holds);
let mut storage_policy_changes = Vec::new();
let mut compute_policy_changes: BTreeMap<_, Vec<_>> = BTreeMap::new();
let new_time = Antichain::from_elem(new_time);
for (old_time, id_bundle) in old_holds {
let new_time = old_time.join(&new_time);
if old_time != new_time {
read_holds
.holds
.entry(new_time.clone())
.or_default()
.extend(&id_bundle);
for id in id_bundle.storage_ids {
// TODO(aljoscha): This is a bit iffy, because we're using
// the since without having a read hold in place. In this
// case it's fine because we are the one controlling the
// read policy and the storage controller currently cannot
// advance its frontiers concurrently.
//
// This will be fixed properly in a future commit that makes
// the storage controller concurrent.
let (read_frontier, _upper) = self
.controller
.storage
.collection_frontiers(id)
.expect("missing storage collection");
assert!(read_frontier.le(&new_time.borrow()),
"Storage collection {:?} has read frontier {:?} not less-equal new time {:?}; old time: {:?}",
id,
read_frontier,
new_time,
old_time,
);
let read_needs = self
.storage_read_capabilities
.get_mut(&id)
.expect("id does not exist");
read_needs
.holds
.update_iter(new_time.iter().map(|t| (*t, 1)));
read_needs
.holds
.update_iter(old_time.iter().map(|t| (*t, -1)));
storage_policy_changes.push((id, read_needs.policy()));
}
for (compute_instance, compute_ids) in id_bundle.compute_ids {
let compute = self.controller.active_compute();
for id in compute_ids {
let collection = compute
.collection(compute_instance, id)
.expect("id does not exist");
assert!(collection.read_capability().le(&new_time.borrow()),
"Compute collection {:?} (instance {:?}) has read frontier {:?} not less-equal new time {:?}; old time: {:?}",
id,
compute_instance,
collection.read_capability(),
new_time,
old_time,
);
let read_needs = self
.compute_read_capabilities
.get_mut(&id)
.expect("id does not exist");
read_needs
.holds
.update_iter(new_time.iter().map(|t| (*t, 1)));
read_needs
.holds
.update_iter(old_time.iter().map(|t| (*t, -1)));
compute_policy_changes
.entry(compute_instance)
.or_default()
.push((id, read_needs.policy()));
}
}
} else {
read_holds
.holds
.entry(old_time)
.or_default()
.extend(&id_bundle);
}
}
// Update STORAGE read policies.
self.controller
.storage
.set_read_policy(storage_policy_changes);
// Update COMPUTE read policies
let mut compute = self.controller.active_compute();
for (compute_instance, compute_policy_changes) in compute_policy_changes {
compute
.set_read_policy(compute_instance, compute_policy_changes)
.unwrap_or_terminate("cannot fail to set read policy");
}
}
/// If there is not capability for the given object, initialize one at the
/// earliest possible since. Return the capability.
//
/// When a `compaction_window` is given, this is installed as the policy of
/// the collection, regardless if a capability existed before or not.
fn ensure_compute_capability(
&mut self,
instance_id: &ComputeInstanceId,
id: &GlobalId,
compaction_window: Option<CompactionWindow>,
) -> &mut ReadCapability<mz_repr::Timestamp> {
let entry = self
.compute_read_capabilities
.entry(*id)
.and_modify(|capability| {
// If we explicitly got a compaction window, override any existing
// one.
if let Some(compaction_window) = compaction_window {
capability.base_policy = compaction_window.into();
}
})
.or_insert_with(|| {
let policy: ReadPolicy<Timestamp> = match compaction_window {
Some(compaction_window) => compaction_window.into(),
None => {
// We didn't get an initial policy, so set the current
// since as a static policy.
let compute = self.controller.active_compute();
let collection = compute
.collection(*instance_id, *id)
.expect("collection does not exist");
let read_frontier = collection.read_capability().clone();
ReadPolicy::ValidFrom(read_frontier)
}
};
ReadCapability::from(policy)
});
entry
}
/// If there is not capability for the given object, initialize one at the
/// earliest possible since. Return the capability.
///
/// When a `compaction_window` is given, this is installed as the policy of
/// the collection, regardless if a capability existed before or not.
fn ensure_storage_capability(
&mut self,
id: &GlobalId,
compaction_window: Option<CompactionWindow>,
) -> &mut ReadCapability<mz_repr::Timestamp> {
let entry = self
.storage_read_capabilities
.entry(*id)
.and_modify(|capability| {
// If we explicitly got a compaction window, override any existing
// one.
if let Some(compaction_window) = compaction_window {
capability.base_policy = compaction_window.into();
}
})
.or_insert_with(|| {
let policy: ReadPolicy<Timestamp> = match compaction_window {
Some(compaction_window) => compaction_window.into(),
None => {
// We didn't get an initial policy, so set the current
// since as a static policy.
//
// N.B. This is a bit iffy, because we're using the
// since without having a read hold in place. In this
// case it's fine because we didn't yet install a
// ReadPolicy at the controller, and the since will stay
// put until we put in place such a policy.
let (since, _upper) = self
.controller
.storage
.collection_frontiers(*id)
.expect("missing storage collection");
ReadPolicy::ValidFrom(since)
}
};
ReadCapability::from(policy)
});
entry
}
pub(crate) fn update_storage_base_read_policies(
&mut self,
base_policies: Vec<(GlobalId, ReadPolicy<mz_repr::Timestamp>)>,
) {
let mut policies = Vec::with_capacity(base_policies.len());
for (id, base_policy) in base_policies {
let capability = self
.storage_read_capabilities
.get_mut(&id)
.expect("coord out of sync");
capability.base_policy = base_policy;
policies.push((id, capability.policy()))
}
self.controller.storage.set_read_policy(policies)
}
pub(crate) fn update_compute_base_read_policies(
&mut self,
mut base_policies: Vec<(ComputeInstanceId, GlobalId, ReadPolicy<mz_repr::Timestamp>)>,
) {
base_policies.sort_by_key(|&(cluster_id, _, _)| cluster_id);
for (cluster_id, group) in &base_policies
.into_iter()
.group_by(|&(cluster_id, _, _)| cluster_id)
{
let group = group
.map(|(_, id, base_policy)| {
let capability = self
.compute_read_capabilities
.get_mut(&id)
.expect("coord out of sync");
capability.base_policy = base_policy;
(id, capability.policy())
})
.collect::<Vec<_>>();
self.controller
.active_compute()
.set_read_policy(cluster_id, group)
.unwrap_or_terminate("cannot fail to set read policy");
}
}
pub(crate) fn update_compute_base_read_policy(
&mut self,
compute_instance: ComputeInstanceId,
id: GlobalId,
base_policy: ReadPolicy<mz_repr::Timestamp>,
) {
self.update_compute_base_read_policies(vec![(compute_instance, id, base_policy)])
}
/// Drop read policy in STORAGE for `id`.
///
/// Returns true if `id` had a read policy and false otherwise.
pub(crate) fn drop_storage_read_policy(&mut self, id: &GlobalId) -> bool {
self.storage_read_capabilities.remove(id).is_some()
}
/// Drop read policy in COMPUTE for `id`.
///
/// Returns true if `id` had a read policy and false otherwise.
pub(crate) fn drop_compute_read_policy(&mut self, id: &GlobalId) -> bool {
self.compute_read_capabilities.remove(id).is_some()
}
/// Attempt to acquire read holds on the indicated collections at the
/// earliest available time.
///
/// # Panics
///
/// Will panic if any of the referenced collections in `id_bundle` don't
/// exist.
pub(crate) fn acquire_read_holds(
&mut self,
id_bundle: &CollectionIdBundle,
) -> ReadHolds<Timestamp> {
// Create a `ReadHoldsInner` that contains a read hold for each id in
// `id_bundle`. The time of each read holds is at `time`, if possible
// otherwise it is at the lowest possible time.
//
// This does not apply the read holds in STORAGE or COMPUTE. The code
// below applies those in the correct read capability.
let mut read_holds = ReadHoldsInner::new();
let time_antichain = Antichain::from_elem(Timestamp::MIN);
for id in id_bundle.storage_ids.iter() {
// TODO(aljoscha): This is a bit iffy, because we're using the since
// without having a read hold in place. In this case it's fine
// because we are the one controlling the read policy and the
// storage controller currently cannot advance its frontiers
// concurrently.
//
// This will be fixed properly in a future commit that makes the
// storage controller concurrent.
let (read_frontier, _upper) = self
.controller
.storage
.collection_frontiers(*id)
.expect("missing storage collection");
let time_antichain = time_antichain.join(&read_frontier);
let hold_chain = MutableAntichain::from(time_antichain);
read_holds.storage_holds.insert(*id, hold_chain);
}
for (compute_instance, compute_ids) in id_bundle.compute_ids.iter() {
let compute = self.controller.active_compute();
for id in compute_ids.iter() {
let collection = compute
.collection(*compute_instance, *id)
.expect("collection does not exist");
let read_frontier = collection.read_capability().clone();
let time_antichain = time_antichain.join(&read_frontier);
let hold_chain = MutableAntichain::from(time_antichain);
read_holds
.compute_holds
.insert((*compute_instance, *id), hold_chain);
}
}
// Update STORAGE read policies.
let mut policy_changes = Vec::new();
for (id, hold) in read_holds.storage_holds.iter_mut() {
let read_needs = self.ensure_storage_capability(id, None);
read_needs.holds.update_iter(hold.updates().cloned());
policy_changes.push((*id, read_needs.policy()));
}
self.controller.storage.set_read_policy(policy_changes);
// Update COMPUTE read policies
let mut policy_changes: HashMap<_, Vec<_>> = HashMap::new();
for ((compute_instance, id), hold) in read_holds.compute_holds.iter_mut() {
let read_needs = self.ensure_compute_capability(compute_instance, id, None);
read_needs.holds.update_iter(hold.updates().cloned());
policy_changes
.entry(*compute_instance)
.or_default()
.push((*id, read_needs.policy()));
}
let mut compute = self.controller.active_compute();
for (compute_instance, policy_changes) in policy_changes {
compute
.set_read_policy(compute_instance, policy_changes)
.unwrap_or_terminate("cannot fail to set read policy");
}
let read_holds = ReadHolds::new(read_holds, self.dropped_read_holds_tx.clone());
read_holds
}
/// Stash transaction read holds. They will be released when the transaction
/// is cleaned up.
pub(crate) fn store_transaction_read_holds(
&mut self,
session: &Session,
read_holds: ReadHolds<Timestamp>,
) {
let entry = self.txn_read_holds.entry(session.conn_id().clone());
match entry {
btree_map::Entry::Vacant(v) => {
v.insert(read_holds);
}
btree_map::Entry::Occupied(mut o) => {
o.get_mut().merge(read_holds);
}
}
}
/// Release the given read holds.
///
/// This method relies on a previous call to
/// `initialize_read_holds`, `acquire_read_holds`, or `update_read_hold` that returned
/// `ReadHolds`, and its behavior will be erratic if called on anything else,
/// or if called more than once on the same bundle of read holds.
pub(super) fn release_read_holds(&mut self, mut read_holdses: Vec<ReadHoldsInner<Timestamp>>) {
// Update STORAGE read policies.
let mut storage_policy_changes = Vec::new();
for read_holds in read_holdses.iter_mut() {
for (id, hold) in read_holds.storage_holds.iter_mut() {
// It's possible that a concurrent DDL statement has already dropped this GlobalId
if let Some(read_needs) = self.storage_read_capabilities.get_mut(id) {
let inverted_hold = hold.updates().map(|(t, diff)| (*t, -diff));
read_needs.holds.update_iter(inverted_hold);
storage_policy_changes.push((*id, read_needs.policy()));
}
}
}
self.controller
.storage
.set_read_policy(storage_policy_changes);
// Update COMPUTE read policies
let mut policy_changes_per_instance = BTreeMap::new();
for read_holds in read_holdses.iter_mut() {
for ((compute_instance, id), hold) in read_holds.compute_holds.iter_mut() {
let policy_changes = policy_changes_per_instance
.entry(compute_instance)
.or_insert_with(Vec::new);
// It's possible that a concurrent DDL statement has already dropped this GlobalId
if let Some(read_needs) = self.compute_read_capabilities.get_mut(id) {
let inverted_hold = hold.updates().map(|(t, diff)| (*t, -diff));
read_needs.holds.update_iter(inverted_hold);
policy_changes.push((*id, read_needs.policy()));
}
}
}
let mut compute = self.controller.active_compute();
for (compute_instance, policy_changes) in policy_changes_per_instance {
if compute.instance_exists(*compute_instance) {
compute
.set_read_policy(*compute_instance, policy_changes)
.unwrap_or_terminate("cannot fail to set read policy");
}
}
}
}