Files
scylladb/utils/logalloc.hh
Benny Halevy e5ca65f78b test/perf: report also log_allocations/op
Currently perf-simple-query --write ignores
log allocations that happen on the memtable
apply path.

This change adds tracking and accounting
of the number of log allocation,
and reporting of thereof.

For reference, here's the output of
build/release/scylla perf-simple-query --write --default-log-level=error --random-seed=1 -c 1
```
random-seed=1
enable-cache=1
Running test with config: {partitions=10000, concurrency=100, mode=write, frontend=cql, query_single_key=no, counters=no}
Disabling auto compaction
78073.55 tps ( 59.4 allocs/op,  16.3 logallocs/op,  14.3 tasks/op,   52991 insns/op,        0 errors)
77263.59 tps ( 59.3 allocs/op,  16.0 logallocs/op,  14.3 tasks/op,   53282 insns/op,        0 errors)
79913.07 tps ( 59.3 allocs/op,  16.0 logallocs/op,  14.3 tasks/op,   53295 insns/op,        0 errors)
79554.32 tps ( 59.3 allocs/op,  16.0 logallocs/op,  14.3 tasks/op,   53284 insns/op,        0 errors)
79151.53 tps ( 59.3 allocs/op,  16.0 logallocs/op,  14.3 tasks/op,   53289 insns/op,        0 errors)

median 79151.53 tps ( 59.3 allocs/op,  16.0 logallocs/op,  14.3 tasks/op,   53289 insns/op,        0 errors)
median absolute deviation: 761.54
maximum: 79913.07
minimum: 77263.59
```

Signed-off-by: Benny Halevy <bhalevy@scylladb.com>
2024-05-02 18:42:41 +03:00

551 lines
18 KiB
C++

/*
* Copyright (C) 2015-present ScyllaDB
*/
/*
* SPDX-License-Identifier: AGPL-3.0-or-later
*/
#pragma once
#include <memory>
#include <seastar/core/memory.hh>
#include <seastar/core/condition-variable.hh>
#include <seastar/core/smp.hh>
#include <seastar/core/shared_ptr.hh>
#include <seastar/core/shared_future.hh>
#include <seastar/core/expiring_fifo.hh>
#include "allocation_strategy.hh"
#include "seastarx.hh"
#include "db/timeout_clock.hh"
#include "utils/entangled.hh"
#include "utils/memory_limit_reached.hh"
namespace logalloc {
struct occupancy_stats;
class region;
class region_impl;
class allocating_section;
constexpr int segment_size_shift = 17; // 128K; see #151, #152
constexpr size_t segment_size = 1 << segment_size_shift;
constexpr size_t max_zone_segments = 256;
//
// Frees some amount of objects from the region to which it's attached.
//
// This should eventually stop given no new objects are added:
//
// while (eviction_fn() == memory::reclaiming_result::reclaimed_something) ;
//
using eviction_fn = std::function<memory::reclaiming_result()>;
// Listens for events from a region
class region_listener {
public:
virtual ~region_listener();
virtual void add(region* r) = 0;
virtual void del(region* r) = 0;
virtual void moved(region* old_address, region* new_address) = 0;
virtual void increase_usage(region* r, ssize_t delta) = 0;
virtual void decrease_evictable_usage(region* r) = 0;
virtual void decrease_usage(region* r, ssize_t delta) = 0;
};
// Controller for all LSA regions. There's one per shard.
class tracker {
public:
class impl;
struct config {
bool defragment_on_idle;
bool abort_on_lsa_bad_alloc;
bool sanitizer_report_backtrace = false; // Better reports but slower
size_t lsa_reclamation_step;
scheduling_group background_reclaim_sched_group;
};
struct stats {
size_t segments_compacted;
size_t lsa_buffer_segments;
uint64_t memory_allocated;
uint64_t memory_freed;
uint64_t memory_compacted;
uint64_t memory_evicted;
uint64_t num_allocations;
friend stats operator+(const stats& s1, const stats& s2) {
stats result(s1);
result += s2;
return result;
}
friend stats operator-(const stats& s1, const stats& s2) {
stats result(s1);
result -= s2;
return result;
}
stats& operator+=(const stats& other) {
segments_compacted += other.segments_compacted;
lsa_buffer_segments += other.lsa_buffer_segments;
memory_allocated += other.memory_allocated;
memory_freed += other.memory_freed;
memory_compacted += other.memory_compacted;
memory_evicted += other.memory_evicted;
num_allocations += other.num_allocations;
return *this;
}
stats& operator-=(const stats& other) {
segments_compacted -= other.segments_compacted;
lsa_buffer_segments -= other.lsa_buffer_segments;
memory_allocated -= other.memory_allocated;
memory_freed -= other.memory_freed;
memory_compacted -= other.memory_compacted;
memory_evicted -= other.memory_evicted;
num_allocations -= other.num_allocations;
return *this;
}
};
void configure(const config& cfg);
future<> stop();
private:
std::unique_ptr<impl> _impl;
memory::reclaimer _reclaimer;
friend class region;
friend class region_impl;
memory::reclaiming_result reclaim(seastar::memory::reclaimer::request);
public:
tracker();
~tracker();
stats statistics() const;
//
// Tries to reclaim given amount of bytes in total using all compactible
// and evictable regions. Returns the number of bytes actually reclaimed.
// That value may be smaller than requested when evictable pools are empty
// and compactible pools can't compact any more.
//
// Invalidates references to objects in all compactible and evictable regions.
//
size_t reclaim(size_t bytes);
// Compacts as much as possible. Very expensive, mainly for testing.
// Guarantees that every live object from reclaimable regions will be moved.
// Invalidates references to objects in all compactible and evictable regions.
void full_compaction();
void reclaim_all_free_segments();
occupancy_stats global_occupancy() const noexcept;
// Returns aggregate statistics for all pools.
occupancy_stats region_occupancy() const noexcept;
// Returns statistics for all segments allocated by LSA on this shard.
occupancy_stats occupancy() const noexcept;
// Returns amount of allocated memory not managed by LSA
size_t non_lsa_used_space() const noexcept;
impl& get_impl() noexcept { return *_impl; }
// Returns the minimum number of segments reclaimed during single reclamation cycle.
size_t reclamation_step() const noexcept;
bool should_abort_on_bad_alloc() const noexcept;
};
class tracker_reclaimer_lock {
tracker::impl& _tracker_impl;
public:
tracker_reclaimer_lock(tracker::impl& impl) noexcept;
tracker_reclaimer_lock(tracker& t) noexcept : tracker_reclaimer_lock(t.get_impl()) { }
~tracker_reclaimer_lock();
};
tracker& shard_tracker() noexcept;
class segment_descriptor;
/// A unique pointer to a chunk of memory allocated inside an LSA region.
///
/// The pointer can be in disengaged state in which case it doesn't point at any buffer (nullptr state).
/// When the pointer points at some buffer, it is said to be engaged.
///
/// The pointer owns the object.
/// When the pointer is destroyed or it transitions from engaged to disengaged state, the buffer is freed.
/// The buffer is never leaked when operating by the API of lsa_buffer.
/// The pointer object can be safely destroyed in any allocator context.
///
/// The pointer object is never invalidated.
/// The pointed-to buffer can be moved around by LSA, so the pointer returned by get() can be
/// invalidated, but the pointer object itself is updated automatically and get() always returns
/// a pointer which is valid at the time of the call.
///
/// Must not outlive the region.
class lsa_buffer {
friend class region_impl;
entangled _link; // Paired with segment_descriptor::_buf_pointers[...]
segment_descriptor* _desc; // Valid only when engaged
char* _buf = nullptr; // Valid only when engaged
size_t _size = 0;
public:
using char_type = char;
lsa_buffer() = default;
lsa_buffer(lsa_buffer&&) noexcept = default;
~lsa_buffer();
/// Makes this instance point to the buffer pointed to by the other pointer.
/// If this pointer was engaged before, the owned buffer is freed.
/// The other pointer will be in disengaged state after this.
lsa_buffer& operator=(lsa_buffer&& other) noexcept {
if (this != &other) {
this->~lsa_buffer();
new (this) lsa_buffer(std::move(other));
}
return *this;
}
/// Disengages the pointer.
/// If the pointer was engaged before, the owned buffer is freed.
/// Postcondition: !bool(*this)
lsa_buffer& operator=(std::nullptr_t) noexcept {
this->~lsa_buffer();
return *this;
}
/// Returns a pointer to the first element of the buffer.
/// Valid only when engaged.
char_type* get() noexcept { return _buf; }
const char_type* get() const noexcept { return _buf; }
/// Returns the number of bytes in the buffer.
size_t size() const noexcept { return _size; }
/// Returns true iff the pointer is engaged.
explicit operator bool() const noexcept { return bool(_link); }
};
// Monoid representing pool occupancy statistics.
// Naturally ordered so that sparser pools come fist.
// All sizes in bytes.
class occupancy_stats {
size_t _free_space;
size_t _total_space;
public:
occupancy_stats() noexcept : _free_space(0), _total_space(0) {}
occupancy_stats(size_t free_space, size_t total_space) noexcept
: _free_space(free_space), _total_space(total_space) { }
bool operator<(const occupancy_stats& other) const noexcept {
return used_fraction() < other.used_fraction();
}
friend occupancy_stats operator+(const occupancy_stats& s1, const occupancy_stats& s2) noexcept {
occupancy_stats result(s1);
result += s2;
return result;
}
friend occupancy_stats operator-(const occupancy_stats& s1, const occupancy_stats& s2) noexcept {
occupancy_stats result(s1);
result -= s2;
return result;
}
occupancy_stats& operator+=(const occupancy_stats& other) noexcept {
_total_space += other._total_space;
_free_space += other._free_space;
return *this;
}
occupancy_stats& operator-=(const occupancy_stats& other) noexcept {
_total_space -= other._total_space;
_free_space -= other._free_space;
return *this;
}
size_t used_space() const noexcept {
return _total_space - _free_space;
}
size_t free_space() const noexcept {
return _free_space;
}
size_t total_space() const noexcept {
return _total_space;
}
float used_fraction() const noexcept {
return _total_space ? float(used_space()) / total_space() : 0;
}
explicit operator bool() const noexcept {
return _total_space > 0;
}
};
class basic_region_impl : public allocation_strategy {
protected:
tracker& _tracker;
bool _reclaiming_enabled = true;
seastar::shard_id _cpu = this_shard_id();
public:
basic_region_impl(tracker& tracker) : _tracker(tracker)
{ }
tracker& get_tracker() { return _tracker; }
void set_reclaiming_enabled(bool enabled) noexcept {
assert(this_shard_id() == _cpu);
_reclaiming_enabled = enabled;
}
bool reclaiming_enabled() const noexcept {
return _reclaiming_enabled;
}
};
//
// Log-structured allocator region.
//
// Objects allocated using this region are said to be owned by this region.
// Objects must be freed only using the region which owns them. Ownership can
// be transferred across regions using the merge() method. Region must be live
// as long as it owns any objects.
//
// Each region has separate memory accounting and can be compacted
// independently from other regions. To reclaim memory from all regions use
// shard_tracker().
//
// Region is automatically added to the set of
// compactible regions when constructed.
//
class region {
public:
using impl = region_impl;
private:
shared_ptr<basic_region_impl> _impl;
private:
region_impl& get_impl() noexcept;
const region_impl& get_impl() const noexcept;
public:
region();
~region();
region(region&& other) noexcept;
region& operator=(region&& other) noexcept;
region(const region& other) = delete;
void listen(region_listener* listener);
void unlisten();
occupancy_stats occupancy() const noexcept;
tracker& get_tracker() const {
return _impl->get_tracker();
}
allocation_strategy& allocator() noexcept {
return *_impl;
}
const allocation_strategy& allocator() const noexcept {
return *_impl;
}
// Allocates a buffer of a given size.
// The buffer's pointer will be aligned to 4KB.
// Note: it is wasteful to allocate buffers of sizes which are not a multiple of the alignment.
lsa_buffer alloc_buf(size_t buffer_size);
// Merges another region into this region. The other region is left empty.
// Doesn't invalidate references to allocated objects.
void merge(region& other) noexcept;
// Compacts everything. Mainly for testing.
// Invalidates references to allocated objects.
void full_compaction();
// Runs eviction function once. Mainly for testing.
memory::reclaiming_result evict_some();
// Changes the reclaimability state of this region. When region is not
// reclaimable, it won't be considered by tracker::reclaim(). By default region is
// reclaimable after construction.
void set_reclaiming_enabled(bool e) noexcept { _impl->set_reclaiming_enabled(e); }
// Returns the reclaimability state of this region.
bool reclaiming_enabled() const noexcept { return _impl->reclaiming_enabled(); }
// Returns a value which is increased when this region is either compacted or
// evicted from, which invalidates references into the region.
// When the value returned by this method doesn't change, references remain valid.
uint64_t reclaim_counter() const noexcept {
return allocator().invalidate_counter();
}
// Will cause subsequent calls to evictable_occupancy() to report empty occupancy.
void ground_evictable_occupancy();
// Follows region's occupancy in the parent region group. Less fine-grained than occupancy().
// After ground_evictable_occupancy() is called returns 0.
occupancy_stats evictable_occupancy() const noexcept;
// Makes this region an evictable region. Supplied function will be called
// when data from this region needs to be evicted in order to reclaim space.
// The function should free some space from this region.
void make_evictable(eviction_fn) noexcept;
const eviction_fn& evictor() const noexcept;
uint64_t id() const noexcept;
std::unordered_map<std::string, uint64_t> collect_stats() const;
friend class allocating_section;
};
// Forces references into the region to remain valid as long as this guard is
// live by disabling compaction and eviction.
// Can be nested.
struct reclaim_lock {
region& _region;
bool _prev;
reclaim_lock(region& r) noexcept
: _region(r)
, _prev(r.reclaiming_enabled())
{
_region.set_reclaiming_enabled(false);
}
~reclaim_lock() {
_region.set_reclaiming_enabled(_prev);
}
};
// Utility for running critical sections which need to lock some region and
// also allocate LSA memory. The object learns from failures how much it
// should reserve up front in order to not cause allocation failures.
class allocating_section {
// Do not decay below these minimal values
static constexpr size_t s_min_lsa_reserve = 1;
static constexpr size_t s_min_std_reserve = 1024;
static constexpr uint64_t s_bytes_per_decay = 10'000'000'000;
static constexpr unsigned s_segments_per_decay = 100'000;
size_t _lsa_reserve = s_min_lsa_reserve; // in segments
size_t _std_reserve = s_min_std_reserve; // in bytes
size_t _minimum_lsa_emergency_reserve = 0;
int64_t _remaining_std_bytes_until_decay = s_bytes_per_decay;
int _remaining_lsa_segments_until_decay = s_segments_per_decay;
private:
struct guard {
tracker::impl& _tracker;
size_t _prev;
explicit guard(tracker::impl& tracker) noexcept;
~guard();
};
void reserve(tracker::impl& tracker);
void maybe_decay_reserve() noexcept;
void on_alloc_failure(logalloc::region&);
public:
void set_lsa_reserve(size_t) noexcept;
void set_std_reserve(size_t) noexcept;
//
// Reserves standard allocator and LSA memory for subsequent operations that
// have to be performed with memory reclamation disabled.
//
// Throws std::bad_alloc when reserves can't be increased to a sufficient level.
//
template<typename Func>
decltype(auto) with_reserve(region& r, Func&& fn) {
auto prev_lsa_reserve = _lsa_reserve;
auto prev_std_reserve = _std_reserve;
try {
guard g(r.get_tracker().get_impl());
_minimum_lsa_emergency_reserve = g._prev;
reserve(r.get_tracker().get_impl());
return fn();
} catch (const std::bad_alloc&) {
// roll-back limits to protect against pathological requests
// preventing future requests from succeeding.
_lsa_reserve = prev_lsa_reserve;
_std_reserve = prev_std_reserve;
throw;
}
}
//
// Invokes func with reclaim_lock on region r. If LSA allocation fails
// inside func it is retried after increasing LSA segment reserve. The
// memory reserves are increased with region lock off allowing for memory
// reclamation to take place in the region.
//
// References in the region are invalidated when allocating section is re-entered
// on allocation failure.
//
// Throws std::bad_alloc when reserves can't be increased to a sufficient level.
//
template<typename Func>
decltype(auto) with_reclaiming_disabled(logalloc::region& r, Func&& fn) {
assert(r.reclaiming_enabled());
maybe_decay_reserve();
while (true) {
try {
logalloc::reclaim_lock _(r);
memory::disable_abort_on_alloc_failure_temporarily dfg;
return fn();
} catch (const utils::memory_limit_reached&) {
// Do not retry, bumping reserves won't help.
// The read reached a memory limit in the semaphore and is being
// terminated.
throw;
} catch (const std::bad_alloc&) {
on_alloc_failure(r);
}
}
}
//
// Reserves standard allocator and LSA memory and
// invokes func with reclaim_lock on region r. If LSA allocation fails
// inside func it is retried after increasing LSA segment reserve. The
// memory reserves are increased with region lock off allowing for memory
// reclamation to take place in the region.
//
// References in the region are invalidated when allocating section is re-entered
// on allocation failure.
//
// Throws std::bad_alloc when reserves can't be increased to a sufficient level.
//
template<typename Func>
decltype(auto) operator()(logalloc::region& r, Func&& func) {
return with_reserve(r, [this, &r, &func] {
return with_reclaiming_disabled(r, func);
});
}
};
future<> prime_segment_pool(size_t available_memory, size_t min_free_memory);
// Use the segment pool appropriate for the standard allocator.
//
// In debug mode, this will use the release standard allocator store.
// Call once, when initializing the application, before any LSA allocation takes place.
future<> use_standard_allocator_segment_pool_backend(size_t available_memory);
}
template <> struct fmt::formatter<logalloc::occupancy_stats> : fmt::formatter<string_view> {
auto format(const logalloc::occupancy_stats& stats, fmt::format_context& ctx) const {
return fmt::format_to(ctx.out(), "{:.2f}%, {:d} / {:d} [B]",
stats.used_fraction() * 100, stats.used_space(), stats.total_space());
}
};