##// END OF EJS Templates
discovery: slowly increase sampling size...
discovery: slowly increase sampling size Some pathological discovery runs can requires many roundtrip. When this happens things can get very slow. To make the algorithm more resilience again such pathological case. We slowly increase the sample size with each roundtrip (+5%). This will have a negligible impact on "normal" discovery with few roundtrips, but a large positive impact of case with many roundtrips. Asking more question per roundtrip helps to reduce the undecided set faster. Instead of reducing the undecided set a linear speed (in the worst case), we reduce it as a guaranteed (small) exponential rate. The data below show this slow ramp up in sample size: round trip | 1 | 5 | 10 | 20 | 50 | 100 | 130 | sample size | 200 | 254 | 321 | 517 | 2 199 | 25 123 | 108 549 | covered nodes | 200 | 1 357 | 2 821 | 7 031 | 42 658 | 524 530 | 2 276 755 | To be a bit more concrete, lets take a very pathological case as an example. We are doing discovery from a copy of Mozilla-try to a more recent version of mozilla-unified. Mozilla-unified heads are unknown to the mozilla-try repo and there are over 1 million "missing" changesets. (the discovery is "local" to avoid network interference) Without this change, the discovery: - last 1858 seconds (31 minutes), - does 1700 round trip, - asking about 340 000 nodes. With this change, the discovery: - last 218 seconds (3 minutes, 38 seconds a -88% improvement), - does 94 round trip (-94%), - asking about 344 211 nodes (+1%). Of course, this is an extreme case (and 3 minutes is still slow). However this give a good example of how this sample size increase act as a safety net catching any bad situations. We could image a steeper increase than 5%. For example 10% would give the following number: round trip | 1 | 5 | 10 | 20 | 50 | 75 | 100 | sample size | 200 | 321 | 514 | 1 326 | 23 060 | 249 812 | 2 706 594 | covered nodes | 200 | 1 541 | 3 690 | 12 671 | 251 871 | 2 746 254 | 29 770 966 | In parallel, it is useful to understand these pathological cases and improve them. However the current change provides a general purpose safety net to smooth the impact of pathological cases. To avoid issue with older http server, the increase in sample size only occurs if the protocol has not limit on command argument size.

File last commit:

r42354:be073355 default
r42546:dbd0fcca default
Show More
dagops.rs
276 lines | 8.8 KiB | application/rls-services+xml | RustLexer
// dagops.rs
//
// Copyright 2019 Georges Racinet <georges.racinet@octobus.net>
//
// This software may be used and distributed according to the terms of the
// GNU General Public License version 2 or any later version.
//! Miscellaneous DAG operations
//!
//! # Terminology
//! - By *relative heads* of a collection of revision numbers (`Revision`),
//! we mean those revisions that have no children among the collection.
//! - Similarly *relative roots* of a collection of `Revision`, we mean
//! those whose parents, if any, don't belong to the collection.
use super::{Graph, GraphError, Revision, NULL_REVISION};
use crate::ancestors::AncestorsIterator;
use std::collections::{BTreeSet, HashSet};
fn remove_parents(
graph: &impl Graph,
rev: Revision,
set: &mut HashSet<Revision>,
) -> Result<(), GraphError> {
for parent in graph.parents(rev)?.iter() {
if *parent != NULL_REVISION {
set.remove(parent);
}
}
Ok(())
}
/// Relative heads out of some revisions, passed as an iterator.
///
/// These heads are defined as those revisions that have no children
/// among those emitted by the iterator.
///
/// # Performance notes
/// Internally, this clones the iterator, and builds a `HashSet` out of it.
///
/// This function takes an `Iterator` instead of `impl IntoIterator` to
/// guarantee that cloning the iterator doesn't result in cloning the full
/// construct it comes from.
pub fn heads<'a>(
graph: &impl Graph,
iter_revs: impl Clone + Iterator<Item = &'a Revision>,
) -> Result<HashSet<Revision>, GraphError> {
let mut heads: HashSet<Revision> = iter_revs.clone().cloned().collect();
heads.remove(&NULL_REVISION);
for rev in iter_revs {
if *rev != NULL_REVISION {
remove_parents(graph, *rev, &mut heads)?;
}
}
Ok(heads)
}
/// Retain in `revs` only its relative heads.
///
/// This is an in-place operation, so that control of the incoming
/// set is left to the caller.
/// - a direct Python binding would probably need to build its own `HashSet`
/// from an incoming iterable, even if its sole purpose is to extract the
/// heads.
/// - a Rust caller can decide whether cloning beforehand is appropriate
///
/// # Performance notes
/// Internally, this function will store a full copy of `revs` in a `Vec`.
pub fn retain_heads(
graph: &impl Graph,
revs: &mut HashSet<Revision>,
) -> Result<(), GraphError> {
revs.remove(&NULL_REVISION);
// we need to construct an iterable copy of revs to avoid itering while
// mutating
let as_vec: Vec<Revision> = revs.iter().cloned().collect();
for rev in as_vec {
if rev != NULL_REVISION {
remove_parents(graph, rev, revs)?;
}
}
Ok(())
}
/// Roots of `revs`, passed as a `HashSet`
///
/// They are returned in arbitrary order
pub fn roots<G: Graph>(
graph: &G,
revs: &HashSet<Revision>,
) -> Result<Vec<Revision>, GraphError> {
let mut roots: Vec<Revision> = Vec::new();
for rev in revs {
if graph
.parents(*rev)?
.iter()
.filter(|p| **p != NULL_REVISION)
.all(|p| !revs.contains(p))
{
roots.push(*rev);
}
}
Ok(roots)
}
/// Compute the topological range between two collections of revisions
///
/// This is equivalent to the revset `<roots>::<heads>`.
///
/// Currently, the given `Graph` has to implement `Clone`, which means
/// actually cloning just a reference-counted Python pointer if
/// it's passed over through `rust-cpython`. This is due to the internal
/// use of `AncestorsIterator`
///
/// # Algorithmic details
///
/// This is a two-pass swipe inspired from what `reachableroots2` from
/// `mercurial.cext.parsers` does to obtain the same results.
///
/// - first, we climb up the DAG from `heads` in topological order, keeping
/// them in the vector `heads_ancestors` vector, and adding any element of
/// `roots` we find among them to the resulting range.
/// - Then, we iterate on that recorded vector so that a revision is always
/// emitted after its parents and add all revisions whose parents are already
/// in the range to the results.
///
/// # Performance notes
///
/// The main difference with the C implementation is that
/// the latter uses a flat array with bit flags, instead of complex structures
/// like `HashSet`, making it faster in most scenarios. In theory, it's
/// possible that the present implementation could be more memory efficient
/// for very large repositories with many branches.
pub fn range(
graph: &(impl Graph + Clone),
roots: impl IntoIterator<Item = Revision>,
heads: impl IntoIterator<Item = Revision>,
) -> Result<BTreeSet<Revision>, GraphError> {
let mut range = BTreeSet::new();
let roots: HashSet<Revision> = roots.into_iter().collect();
let min_root: Revision = match roots.iter().cloned().min() {
None => {
return Ok(range);
}
Some(r) => r,
};
// Internally, AncestorsIterator currently maintains a `HashSet`
// of all seen revision, which is also what we record, albeit in an ordered
// way. There's room for improvement on this duplication.
let ait = AncestorsIterator::new(graph.clone(), heads, min_root, true)?;
let mut heads_ancestors: Vec<Revision> = Vec::new();
for revres in ait {
let rev = revres?;
if roots.contains(&rev) {
range.insert(rev);
}
heads_ancestors.push(rev);
}
for rev in heads_ancestors.into_iter().rev() {
for parent in graph.parents(rev)?.iter() {
if *parent != NULL_REVISION && range.contains(parent) {
range.insert(rev);
}
}
}
Ok(range)
}
#[cfg(test)]
mod tests {
use super::*;
use crate::testing::SampleGraph;
/// Apply `retain_heads()` to the given slice and return as a sorted `Vec`
fn retain_heads_sorted(
graph: &impl Graph,
revs: &[Revision],
) -> Result<Vec<Revision>, GraphError> {
let mut revs: HashSet<Revision> = revs.iter().cloned().collect();
retain_heads(graph, &mut revs)?;
let mut as_vec: Vec<Revision> = revs.iter().cloned().collect();
as_vec.sort();
Ok(as_vec)
}
#[test]
fn test_retain_heads() -> Result<(), GraphError> {
assert_eq!(retain_heads_sorted(&SampleGraph, &[4, 5, 6])?, vec![5, 6]);
assert_eq!(
retain_heads_sorted(&SampleGraph, &[4, 1, 6, 12, 0])?,
vec![1, 6, 12]
);
assert_eq!(
retain_heads_sorted(&SampleGraph, &[1, 2, 3, 4, 5, 6, 7, 8, 9])?,
vec![3, 5, 8, 9]
);
Ok(())
}
/// Apply `heads()` to the given slice and return as a sorted `Vec`
fn heads_sorted(
graph: &impl Graph,
revs: &[Revision],
) -> Result<Vec<Revision>, GraphError> {
let heads = heads(graph, revs.iter())?;
let mut as_vec: Vec<Revision> = heads.iter().cloned().collect();
as_vec.sort();
Ok(as_vec)
}
#[test]
fn test_heads() -> Result<(), GraphError> {
assert_eq!(heads_sorted(&SampleGraph, &[4, 5, 6])?, vec![5, 6]);
assert_eq!(
heads_sorted(&SampleGraph, &[4, 1, 6, 12, 0])?,
vec![1, 6, 12]
);
assert_eq!(
heads_sorted(&SampleGraph, &[1, 2, 3, 4, 5, 6, 7, 8, 9])?,
vec![3, 5, 8, 9]
);
Ok(())
}
/// Apply `roots()` and sort the result for easier comparison
fn roots_sorted(
graph: &impl Graph,
revs: &[Revision],
) -> Result<Vec<Revision>, GraphError> {
let mut as_vec = roots(graph, &revs.iter().cloned().collect())?;
as_vec.sort();
Ok(as_vec)
}
#[test]
fn test_roots() -> Result<(), GraphError> {
assert_eq!(roots_sorted(&SampleGraph, &[4, 5, 6])?, vec![4]);
assert_eq!(
roots_sorted(&SampleGraph, &[4, 1, 6, 12, 0])?,
vec![0, 4, 12]
);
assert_eq!(
roots_sorted(&SampleGraph, &[1, 2, 3, 4, 5, 6, 7, 8, 9])?,
vec![1, 8]
);
Ok(())
}
/// Apply `range()` and convert the result into a Vec for easier comparison
fn range_vec(
graph: impl Graph + Clone,
roots: &[Revision],
heads: &[Revision],
) -> Result<Vec<Revision>, GraphError> {
range(&graph, roots.iter().cloned(), heads.iter().cloned())
.map(|bs| bs.into_iter().collect())
}
#[test]
fn test_range() -> Result<(), GraphError> {
assert_eq!(range_vec(SampleGraph, &[0], &[4])?, vec![0, 1, 2, 4]);
assert_eq!(range_vec(SampleGraph, &[0], &[8])?, vec![]);
assert_eq!(
range_vec(SampleGraph, &[5, 6], &[10, 11, 13])?,
vec![5, 10]
);
assert_eq!(
range_vec(SampleGraph, &[5, 6], &[10, 12])?,
vec![5, 6, 9, 10, 12]
);
Ok(())
}
}