rocksdb/db/db_iter_stress_test.cc

655 lines
21 KiB
C++
Raw Normal View History

Change and clarify the relationship between Valid(), status() and Seek*() for all iterators. Also fix some bugs Summary: Before this PR, Iterator/InternalIterator may simultaneously have non-ok status() and Valid() = true. That state means that the last operation failed, but the iterator is nevertheless positioned on some unspecified record. Likely intended uses of that are: * If some sst files are corrupted, a normal iterator can be used to read the data from files that are not corrupted. * When using read_tier = kBlockCacheTier, read the data that's in block cache, skipping over the data that is not. However, this behavior wasn't documented well (and until recently the wiki on github had misleading incorrect information). In the code there's a lot of confusion about the relationship between status() and Valid(), and about whether Seek()/SeekToLast()/etc reset the status or not. There were a number of bugs caused by this confusion, both inside rocksdb and in the code that uses rocksdb (including ours). This PR changes the convention to: * If status() is not ok, Valid() always returns false. * Any seek operation resets status. (Before the PR, it depended on iterator type and on particular error.) This does sacrifice the two use cases listed above, but siying said it's ok. Overview of the changes: * A commit that adds missing status checks in MergingIterator. This fixes a bug that actually affects us, and we need it fixed. `DBIteratorTest.NonBlockingIterationBugRepro` explains the scenario. * Changes to lots of iterator types to make all of them conform to the new convention. Some bug fixes along the way. By far the biggest changes are in DBIter, which is a big messy piece of code; I tried to make it less big and messy but mostly failed. * A stress-test for DBIter, to gain some confidence that I didn't break it. It does a few million random operations on the iterator, while occasionally modifying the underlying data (like ForwardIterator does) and occasionally returning non-ok status from internal iterator. To find the iterator types that needed changes I searched for "public .*Iterator" in the code. Here's an overview of all 27 iterator types: Iterators that didn't need changes: * status() is always ok(), or Valid() is always false: MemTableIterator, ModelIter, TestIterator, KVIter (2 classes with this name anonymous namespaces), LoggingForwardVectorIterator, VectorIterator, MockTableIterator, EmptyIterator, EmptyInternalIterator. * Thin wrappers that always pass through Valid() and status(): ArenaWrappedDBIter, TtlIterator, InternalIteratorFromIterator. Iterators with changes (see inline comments for details): * DBIter - an overhaul: - It used to silently skip corrupted keys (`FindParseableKey()`), which seems dangerous. This PR makes it just stop immediately after encountering a corrupted key, just like it would for other kinds of corruption. Let me know if there was actually some deeper meaning in this behavior and I should put it back. - It had a few code paths silently discarding subiterator's status. The stress test caught a few. - The backwards iteration code path was expecting the internal iterator's set of keys to be immutable. It's probably always true in practice at the moment, since ForwardIterator doesn't support backwards iteration, but this PR fixes it anyway. See added DBIteratorTest.ReverseToForwardBug for an example. - Some parts of backwards iteration code path even did things like `assert(iter_->Valid())` after a seek, which is never a safe assumption. - It used to not reset status on seek for some types of errors. - Some simplifications and better comments. - Some things got more complicated from the added error handling. I'm open to ideas for how to make it nicer. * MergingIterator - check status after every operation on every subiterator, and in some places assert that valid subiterators have ok status. * ForwardIterator - changed to the new convention, also slightly simplified. * ForwardLevelIterator - fixed some bugs and simplified. * LevelIterator - simplified. * TwoLevelIterator - changed to the new convention. Also fixed a bug that would make SeekForPrev() sometimes silently ignore errors from first_level_iter_. * BlockBasedTableIterator - minor changes. * BlockIter - replaced `SetStatus()` with `Invalidate()` to make sure non-ok BlockIter is always invalid. * PlainTableIterator - some seeks used to not reset status. * CuckooTableIterator - tiny code cleanup. * ManagedIterator - fixed some bugs. * BaseDeltaIterator - changed to the new convention and fixed a bug. * BlobDBIterator - seeks used to not reset status. * KeyConvertingIterator - some small change. Closes https://github.com/facebook/rocksdb/pull/3810 Differential Revision: D7888019 Pulled By: al13n321 fbshipit-source-id: 4aaf6d3421c545d16722a815b2fa2e7912bc851d
2018-05-17 09:44:14 +00:00
// Copyright (c) 2011-present, Facebook, Inc. All rights reserved.
// This source code is licensed under both the GPLv2 (found in the
// COPYING file in the root directory) and Apache 2.0 License
// (found in the LICENSE.Apache file in the root directory).
#include "db/db_iter.h"
#include "db/dbformat.h"
#include "rocksdb/comparator.h"
#include "rocksdb/options.h"
#include "rocksdb/slice.h"
#include "test_util/testharness.h"
Change and clarify the relationship between Valid(), status() and Seek*() for all iterators. Also fix some bugs Summary: Before this PR, Iterator/InternalIterator may simultaneously have non-ok status() and Valid() = true. That state means that the last operation failed, but the iterator is nevertheless positioned on some unspecified record. Likely intended uses of that are: * If some sst files are corrupted, a normal iterator can be used to read the data from files that are not corrupted. * When using read_tier = kBlockCacheTier, read the data that's in block cache, skipping over the data that is not. However, this behavior wasn't documented well (and until recently the wiki on github had misleading incorrect information). In the code there's a lot of confusion about the relationship between status() and Valid(), and about whether Seek()/SeekToLast()/etc reset the status or not. There were a number of bugs caused by this confusion, both inside rocksdb and in the code that uses rocksdb (including ours). This PR changes the convention to: * If status() is not ok, Valid() always returns false. * Any seek operation resets status. (Before the PR, it depended on iterator type and on particular error.) This does sacrifice the two use cases listed above, but siying said it's ok. Overview of the changes: * A commit that adds missing status checks in MergingIterator. This fixes a bug that actually affects us, and we need it fixed. `DBIteratorTest.NonBlockingIterationBugRepro` explains the scenario. * Changes to lots of iterator types to make all of them conform to the new convention. Some bug fixes along the way. By far the biggest changes are in DBIter, which is a big messy piece of code; I tried to make it less big and messy but mostly failed. * A stress-test for DBIter, to gain some confidence that I didn't break it. It does a few million random operations on the iterator, while occasionally modifying the underlying data (like ForwardIterator does) and occasionally returning non-ok status from internal iterator. To find the iterator types that needed changes I searched for "public .*Iterator" in the code. Here's an overview of all 27 iterator types: Iterators that didn't need changes: * status() is always ok(), or Valid() is always false: MemTableIterator, ModelIter, TestIterator, KVIter (2 classes with this name anonymous namespaces), LoggingForwardVectorIterator, VectorIterator, MockTableIterator, EmptyIterator, EmptyInternalIterator. * Thin wrappers that always pass through Valid() and status(): ArenaWrappedDBIter, TtlIterator, InternalIteratorFromIterator. Iterators with changes (see inline comments for details): * DBIter - an overhaul: - It used to silently skip corrupted keys (`FindParseableKey()`), which seems dangerous. This PR makes it just stop immediately after encountering a corrupted key, just like it would for other kinds of corruption. Let me know if there was actually some deeper meaning in this behavior and I should put it back. - It had a few code paths silently discarding subiterator's status. The stress test caught a few. - The backwards iteration code path was expecting the internal iterator's set of keys to be immutable. It's probably always true in practice at the moment, since ForwardIterator doesn't support backwards iteration, but this PR fixes it anyway. See added DBIteratorTest.ReverseToForwardBug for an example. - Some parts of backwards iteration code path even did things like `assert(iter_->Valid())` after a seek, which is never a safe assumption. - It used to not reset status on seek for some types of errors. - Some simplifications and better comments. - Some things got more complicated from the added error handling. I'm open to ideas for how to make it nicer. * MergingIterator - check status after every operation on every subiterator, and in some places assert that valid subiterators have ok status. * ForwardIterator - changed to the new convention, also slightly simplified. * ForwardLevelIterator - fixed some bugs and simplified. * LevelIterator - simplified. * TwoLevelIterator - changed to the new convention. Also fixed a bug that would make SeekForPrev() sometimes silently ignore errors from first_level_iter_. * BlockBasedTableIterator - minor changes. * BlockIter - replaced `SetStatus()` with `Invalidate()` to make sure non-ok BlockIter is always invalid. * PlainTableIterator - some seeks used to not reset status. * CuckooTableIterator - tiny code cleanup. * ManagedIterator - fixed some bugs. * BaseDeltaIterator - changed to the new convention and fixed a bug. * BlobDBIterator - seeks used to not reset status. * KeyConvertingIterator - some small change. Closes https://github.com/facebook/rocksdb/pull/3810 Differential Revision: D7888019 Pulled By: al13n321 fbshipit-source-id: 4aaf6d3421c545d16722a815b2fa2e7912bc851d
2018-05-17 09:44:14 +00:00
#include "util/random.h"
#include "util/string_util.h"
#include "utilities/merge_operators.h"
#ifdef GFLAGS
#include "util/gflags_compat.h"
using GFLAGS_NAMESPACE::ParseCommandLineFlags;
DEFINE_bool(verbose, false,
"Print huge, detailed trace. Intended for debugging failures.");
#else
void ParseCommandLineFlags(int*, char***, bool) {}
bool FLAGS_verbose = false;
#endif
namespace rocksdb {
class DBIteratorStressTest : public testing::Test {
public:
Env* env_;
DBIteratorStressTest() : env_(Env::Default()) {}
};
namespace {
struct Entry {
std::string key;
ValueType type; // kTypeValue, kTypeDeletion, kTypeMerge
uint64_t sequence;
std::string ikey; // internal key, made from `key`, `sequence` and `type`
std::string value;
// If false, we'll pretend that this entry doesn't exist.
bool visible = true;
bool operator<(const Entry& e) const {
if (key != e.key) return key < e.key;
return std::tie(sequence, type) > std::tie(e.sequence, e.type);
}
};
struct Data {
std::vector<Entry> entries;
// Indices in `entries` with `visible` = false.
std::vector<size_t> hidden;
// Keys of entries whose `visible` changed since the last seek of iterators.
std::set<std::string> recently_touched_keys;
};
struct StressTestIterator : public InternalIterator {
Data* data;
Random64* rnd;
InternalKeyComparator cmp;
// Each operation will return error with this probability...
double error_probability = 0;
// ... and add/remove entries with this probability.
double mutation_probability = 0;
// The probability of adding vs removing entries will be chosen so that the
// amount of removed entries stays somewhat close to this number.
double target_hidden_fraction = 0;
// If true, print all mutations to stdout for debugging.
bool trace = false;
int iter = -1;
Status status_;
StressTestIterator(Data* _data, Random64* _rnd, const Comparator* _cmp)
: data(_data), rnd(_rnd), cmp(_cmp) {}
bool Valid() const override {
if (iter >= 0 && iter < (int)data->entries.size()) {
assert(status_.ok());
return true;
}
return false;
}
Status status() const override { return status_; }
bool MaybeFail() {
if (rnd->Next() >=
std::numeric_limits<uint64_t>::max() * error_probability) {
return false;
}
if (rnd->Next() % 2) {
status_ = Status::Incomplete("test");
} else {
status_ = Status::IOError("test");
}
if (trace) {
std::cout << "injecting " << status_.ToString() << std::endl;
}
iter = -1;
return true;
}
void MaybeMutate() {
if (rnd->Next() >=
std::numeric_limits<uint64_t>::max() * mutation_probability) {
return;
}
do {
// If too many entries are hidden, hide less, otherwise hide more.
double hide_probability =
data->hidden.size() > data->entries.size() * target_hidden_fraction
? 1. / 3
: 2. / 3;
if (data->hidden.empty()) {
hide_probability = 1;
}
bool do_hide =
rnd->Next() < std::numeric_limits<uint64_t>::max() * hide_probability;
if (do_hide) {
// Hide a random entry.
size_t idx = rnd->Next() % data->entries.size();
Entry& e = data->entries[idx];
if (e.visible) {
if (trace) {
std::cout << "hiding idx " << idx << std::endl;
}
e.visible = false;
data->hidden.push_back(idx);
data->recently_touched_keys.insert(e.key);
} else {
// Already hidden. Let's go unhide something instead, just because
// it's easy and it doesn't really matter what we do.
do_hide = false;
}
}
if (!do_hide) {
// Unhide a random entry.
size_t hi = rnd->Next() % data->hidden.size();
size_t idx = data->hidden[hi];
if (trace) {
std::cout << "unhiding idx " << idx << std::endl;
}
Entry& e = data->entries[idx];
assert(!e.visible);
e.visible = true;
data->hidden[hi] = data->hidden.back();
data->hidden.pop_back();
data->recently_touched_keys.insert(e.key);
}
} while (rnd->Next() % 3 != 0); // do 3 mutations on average
}
void SkipForward() {
while (iter < (int)data->entries.size() && !data->entries[iter].visible) {
++iter;
}
}
void SkipBackward() {
while (iter >= 0 && !data->entries[iter].visible) {
--iter;
}
}
void SeekToFirst() override {
Change and clarify the relationship between Valid(), status() and Seek*() for all iterators. Also fix some bugs Summary: Before this PR, Iterator/InternalIterator may simultaneously have non-ok status() and Valid() = true. That state means that the last operation failed, but the iterator is nevertheless positioned on some unspecified record. Likely intended uses of that are: * If some sst files are corrupted, a normal iterator can be used to read the data from files that are not corrupted. * When using read_tier = kBlockCacheTier, read the data that's in block cache, skipping over the data that is not. However, this behavior wasn't documented well (and until recently the wiki on github had misleading incorrect information). In the code there's a lot of confusion about the relationship between status() and Valid(), and about whether Seek()/SeekToLast()/etc reset the status or not. There were a number of bugs caused by this confusion, both inside rocksdb and in the code that uses rocksdb (including ours). This PR changes the convention to: * If status() is not ok, Valid() always returns false. * Any seek operation resets status. (Before the PR, it depended on iterator type and on particular error.) This does sacrifice the two use cases listed above, but siying said it's ok. Overview of the changes: * A commit that adds missing status checks in MergingIterator. This fixes a bug that actually affects us, and we need it fixed. `DBIteratorTest.NonBlockingIterationBugRepro` explains the scenario. * Changes to lots of iterator types to make all of them conform to the new convention. Some bug fixes along the way. By far the biggest changes are in DBIter, which is a big messy piece of code; I tried to make it less big and messy but mostly failed. * A stress-test for DBIter, to gain some confidence that I didn't break it. It does a few million random operations on the iterator, while occasionally modifying the underlying data (like ForwardIterator does) and occasionally returning non-ok status from internal iterator. To find the iterator types that needed changes I searched for "public .*Iterator" in the code. Here's an overview of all 27 iterator types: Iterators that didn't need changes: * status() is always ok(), or Valid() is always false: MemTableIterator, ModelIter, TestIterator, KVIter (2 classes with this name anonymous namespaces), LoggingForwardVectorIterator, VectorIterator, MockTableIterator, EmptyIterator, EmptyInternalIterator. * Thin wrappers that always pass through Valid() and status(): ArenaWrappedDBIter, TtlIterator, InternalIteratorFromIterator. Iterators with changes (see inline comments for details): * DBIter - an overhaul: - It used to silently skip corrupted keys (`FindParseableKey()`), which seems dangerous. This PR makes it just stop immediately after encountering a corrupted key, just like it would for other kinds of corruption. Let me know if there was actually some deeper meaning in this behavior and I should put it back. - It had a few code paths silently discarding subiterator's status. The stress test caught a few. - The backwards iteration code path was expecting the internal iterator's set of keys to be immutable. It's probably always true in practice at the moment, since ForwardIterator doesn't support backwards iteration, but this PR fixes it anyway. See added DBIteratorTest.ReverseToForwardBug for an example. - Some parts of backwards iteration code path even did things like `assert(iter_->Valid())` after a seek, which is never a safe assumption. - It used to not reset status on seek for some types of errors. - Some simplifications and better comments. - Some things got more complicated from the added error handling. I'm open to ideas for how to make it nicer. * MergingIterator - check status after every operation on every subiterator, and in some places assert that valid subiterators have ok status. * ForwardIterator - changed to the new convention, also slightly simplified. * ForwardLevelIterator - fixed some bugs and simplified. * LevelIterator - simplified. * TwoLevelIterator - changed to the new convention. Also fixed a bug that would make SeekForPrev() sometimes silently ignore errors from first_level_iter_. * BlockBasedTableIterator - minor changes. * BlockIter - replaced `SetStatus()` with `Invalidate()` to make sure non-ok BlockIter is always invalid. * PlainTableIterator - some seeks used to not reset status. * CuckooTableIterator - tiny code cleanup. * ManagedIterator - fixed some bugs. * BaseDeltaIterator - changed to the new convention and fixed a bug. * BlobDBIterator - seeks used to not reset status. * KeyConvertingIterator - some small change. Closes https://github.com/facebook/rocksdb/pull/3810 Differential Revision: D7888019 Pulled By: al13n321 fbshipit-source-id: 4aaf6d3421c545d16722a815b2fa2e7912bc851d
2018-05-17 09:44:14 +00:00
if (MaybeFail()) return;
MaybeMutate();
status_ = Status::OK();
iter = 0;
SkipForward();
}
void SeekToLast() override {
Change and clarify the relationship between Valid(), status() and Seek*() for all iterators. Also fix some bugs Summary: Before this PR, Iterator/InternalIterator may simultaneously have non-ok status() and Valid() = true. That state means that the last operation failed, but the iterator is nevertheless positioned on some unspecified record. Likely intended uses of that are: * If some sst files are corrupted, a normal iterator can be used to read the data from files that are not corrupted. * When using read_tier = kBlockCacheTier, read the data that's in block cache, skipping over the data that is not. However, this behavior wasn't documented well (and until recently the wiki on github had misleading incorrect information). In the code there's a lot of confusion about the relationship between status() and Valid(), and about whether Seek()/SeekToLast()/etc reset the status or not. There were a number of bugs caused by this confusion, both inside rocksdb and in the code that uses rocksdb (including ours). This PR changes the convention to: * If status() is not ok, Valid() always returns false. * Any seek operation resets status. (Before the PR, it depended on iterator type and on particular error.) This does sacrifice the two use cases listed above, but siying said it's ok. Overview of the changes: * A commit that adds missing status checks in MergingIterator. This fixes a bug that actually affects us, and we need it fixed. `DBIteratorTest.NonBlockingIterationBugRepro` explains the scenario. * Changes to lots of iterator types to make all of them conform to the new convention. Some bug fixes along the way. By far the biggest changes are in DBIter, which is a big messy piece of code; I tried to make it less big and messy but mostly failed. * A stress-test for DBIter, to gain some confidence that I didn't break it. It does a few million random operations on the iterator, while occasionally modifying the underlying data (like ForwardIterator does) and occasionally returning non-ok status from internal iterator. To find the iterator types that needed changes I searched for "public .*Iterator" in the code. Here's an overview of all 27 iterator types: Iterators that didn't need changes: * status() is always ok(), or Valid() is always false: MemTableIterator, ModelIter, TestIterator, KVIter (2 classes with this name anonymous namespaces), LoggingForwardVectorIterator, VectorIterator, MockTableIterator, EmptyIterator, EmptyInternalIterator. * Thin wrappers that always pass through Valid() and status(): ArenaWrappedDBIter, TtlIterator, InternalIteratorFromIterator. Iterators with changes (see inline comments for details): * DBIter - an overhaul: - It used to silently skip corrupted keys (`FindParseableKey()`), which seems dangerous. This PR makes it just stop immediately after encountering a corrupted key, just like it would for other kinds of corruption. Let me know if there was actually some deeper meaning in this behavior and I should put it back. - It had a few code paths silently discarding subiterator's status. The stress test caught a few. - The backwards iteration code path was expecting the internal iterator's set of keys to be immutable. It's probably always true in practice at the moment, since ForwardIterator doesn't support backwards iteration, but this PR fixes it anyway. See added DBIteratorTest.ReverseToForwardBug for an example. - Some parts of backwards iteration code path even did things like `assert(iter_->Valid())` after a seek, which is never a safe assumption. - It used to not reset status on seek for some types of errors. - Some simplifications and better comments. - Some things got more complicated from the added error handling. I'm open to ideas for how to make it nicer. * MergingIterator - check status after every operation on every subiterator, and in some places assert that valid subiterators have ok status. * ForwardIterator - changed to the new convention, also slightly simplified. * ForwardLevelIterator - fixed some bugs and simplified. * LevelIterator - simplified. * TwoLevelIterator - changed to the new convention. Also fixed a bug that would make SeekForPrev() sometimes silently ignore errors from first_level_iter_. * BlockBasedTableIterator - minor changes. * BlockIter - replaced `SetStatus()` with `Invalidate()` to make sure non-ok BlockIter is always invalid. * PlainTableIterator - some seeks used to not reset status. * CuckooTableIterator - tiny code cleanup. * ManagedIterator - fixed some bugs. * BaseDeltaIterator - changed to the new convention and fixed a bug. * BlobDBIterator - seeks used to not reset status. * KeyConvertingIterator - some small change. Closes https://github.com/facebook/rocksdb/pull/3810 Differential Revision: D7888019 Pulled By: al13n321 fbshipit-source-id: 4aaf6d3421c545d16722a815b2fa2e7912bc851d
2018-05-17 09:44:14 +00:00
if (MaybeFail()) return;
MaybeMutate();
status_ = Status::OK();
iter = (int)data->entries.size() - 1;
SkipBackward();
}
void Seek(const Slice& target) override {
Change and clarify the relationship between Valid(), status() and Seek*() for all iterators. Also fix some bugs Summary: Before this PR, Iterator/InternalIterator may simultaneously have non-ok status() and Valid() = true. That state means that the last operation failed, but the iterator is nevertheless positioned on some unspecified record. Likely intended uses of that are: * If some sst files are corrupted, a normal iterator can be used to read the data from files that are not corrupted. * When using read_tier = kBlockCacheTier, read the data that's in block cache, skipping over the data that is not. However, this behavior wasn't documented well (and until recently the wiki on github had misleading incorrect information). In the code there's a lot of confusion about the relationship between status() and Valid(), and about whether Seek()/SeekToLast()/etc reset the status or not. There were a number of bugs caused by this confusion, both inside rocksdb and in the code that uses rocksdb (including ours). This PR changes the convention to: * If status() is not ok, Valid() always returns false. * Any seek operation resets status. (Before the PR, it depended on iterator type and on particular error.) This does sacrifice the two use cases listed above, but siying said it's ok. Overview of the changes: * A commit that adds missing status checks in MergingIterator. This fixes a bug that actually affects us, and we need it fixed. `DBIteratorTest.NonBlockingIterationBugRepro` explains the scenario. * Changes to lots of iterator types to make all of them conform to the new convention. Some bug fixes along the way. By far the biggest changes are in DBIter, which is a big messy piece of code; I tried to make it less big and messy but mostly failed. * A stress-test for DBIter, to gain some confidence that I didn't break it. It does a few million random operations on the iterator, while occasionally modifying the underlying data (like ForwardIterator does) and occasionally returning non-ok status from internal iterator. To find the iterator types that needed changes I searched for "public .*Iterator" in the code. Here's an overview of all 27 iterator types: Iterators that didn't need changes: * status() is always ok(), or Valid() is always false: MemTableIterator, ModelIter, TestIterator, KVIter (2 classes with this name anonymous namespaces), LoggingForwardVectorIterator, VectorIterator, MockTableIterator, EmptyIterator, EmptyInternalIterator. * Thin wrappers that always pass through Valid() and status(): ArenaWrappedDBIter, TtlIterator, InternalIteratorFromIterator. Iterators with changes (see inline comments for details): * DBIter - an overhaul: - It used to silently skip corrupted keys (`FindParseableKey()`), which seems dangerous. This PR makes it just stop immediately after encountering a corrupted key, just like it would for other kinds of corruption. Let me know if there was actually some deeper meaning in this behavior and I should put it back. - It had a few code paths silently discarding subiterator's status. The stress test caught a few. - The backwards iteration code path was expecting the internal iterator's set of keys to be immutable. It's probably always true in practice at the moment, since ForwardIterator doesn't support backwards iteration, but this PR fixes it anyway. See added DBIteratorTest.ReverseToForwardBug for an example. - Some parts of backwards iteration code path even did things like `assert(iter_->Valid())` after a seek, which is never a safe assumption. - It used to not reset status on seek for some types of errors. - Some simplifications and better comments. - Some things got more complicated from the added error handling. I'm open to ideas for how to make it nicer. * MergingIterator - check status after every operation on every subiterator, and in some places assert that valid subiterators have ok status. * ForwardIterator - changed to the new convention, also slightly simplified. * ForwardLevelIterator - fixed some bugs and simplified. * LevelIterator - simplified. * TwoLevelIterator - changed to the new convention. Also fixed a bug that would make SeekForPrev() sometimes silently ignore errors from first_level_iter_. * BlockBasedTableIterator - minor changes. * BlockIter - replaced `SetStatus()` with `Invalidate()` to make sure non-ok BlockIter is always invalid. * PlainTableIterator - some seeks used to not reset status. * CuckooTableIterator - tiny code cleanup. * ManagedIterator - fixed some bugs. * BaseDeltaIterator - changed to the new convention and fixed a bug. * BlobDBIterator - seeks used to not reset status. * KeyConvertingIterator - some small change. Closes https://github.com/facebook/rocksdb/pull/3810 Differential Revision: D7888019 Pulled By: al13n321 fbshipit-source-id: 4aaf6d3421c545d16722a815b2fa2e7912bc851d
2018-05-17 09:44:14 +00:00
if (MaybeFail()) return;
MaybeMutate();
status_ = Status::OK();
// Binary search.
auto it = std::partition_point(
data->entries.begin(), data->entries.end(),
[&](const Entry& e) { return cmp.Compare(e.ikey, target) < 0; });
iter = (int)(it - data->entries.begin());
SkipForward();
}
void SeekForPrev(const Slice& target) override {
Change and clarify the relationship between Valid(), status() and Seek*() for all iterators. Also fix some bugs Summary: Before this PR, Iterator/InternalIterator may simultaneously have non-ok status() and Valid() = true. That state means that the last operation failed, but the iterator is nevertheless positioned on some unspecified record. Likely intended uses of that are: * If some sst files are corrupted, a normal iterator can be used to read the data from files that are not corrupted. * When using read_tier = kBlockCacheTier, read the data that's in block cache, skipping over the data that is not. However, this behavior wasn't documented well (and until recently the wiki on github had misleading incorrect information). In the code there's a lot of confusion about the relationship between status() and Valid(), and about whether Seek()/SeekToLast()/etc reset the status or not. There were a number of bugs caused by this confusion, both inside rocksdb and in the code that uses rocksdb (including ours). This PR changes the convention to: * If status() is not ok, Valid() always returns false. * Any seek operation resets status. (Before the PR, it depended on iterator type and on particular error.) This does sacrifice the two use cases listed above, but siying said it's ok. Overview of the changes: * A commit that adds missing status checks in MergingIterator. This fixes a bug that actually affects us, and we need it fixed. `DBIteratorTest.NonBlockingIterationBugRepro` explains the scenario. * Changes to lots of iterator types to make all of them conform to the new convention. Some bug fixes along the way. By far the biggest changes are in DBIter, which is a big messy piece of code; I tried to make it less big and messy but mostly failed. * A stress-test for DBIter, to gain some confidence that I didn't break it. It does a few million random operations on the iterator, while occasionally modifying the underlying data (like ForwardIterator does) and occasionally returning non-ok status from internal iterator. To find the iterator types that needed changes I searched for "public .*Iterator" in the code. Here's an overview of all 27 iterator types: Iterators that didn't need changes: * status() is always ok(), or Valid() is always false: MemTableIterator, ModelIter, TestIterator, KVIter (2 classes with this name anonymous namespaces), LoggingForwardVectorIterator, VectorIterator, MockTableIterator, EmptyIterator, EmptyInternalIterator. * Thin wrappers that always pass through Valid() and status(): ArenaWrappedDBIter, TtlIterator, InternalIteratorFromIterator. Iterators with changes (see inline comments for details): * DBIter - an overhaul: - It used to silently skip corrupted keys (`FindParseableKey()`), which seems dangerous. This PR makes it just stop immediately after encountering a corrupted key, just like it would for other kinds of corruption. Let me know if there was actually some deeper meaning in this behavior and I should put it back. - It had a few code paths silently discarding subiterator's status. The stress test caught a few. - The backwards iteration code path was expecting the internal iterator's set of keys to be immutable. It's probably always true in practice at the moment, since ForwardIterator doesn't support backwards iteration, but this PR fixes it anyway. See added DBIteratorTest.ReverseToForwardBug for an example. - Some parts of backwards iteration code path even did things like `assert(iter_->Valid())` after a seek, which is never a safe assumption. - It used to not reset status on seek for some types of errors. - Some simplifications and better comments. - Some things got more complicated from the added error handling. I'm open to ideas for how to make it nicer. * MergingIterator - check status after every operation on every subiterator, and in some places assert that valid subiterators have ok status. * ForwardIterator - changed to the new convention, also slightly simplified. * ForwardLevelIterator - fixed some bugs and simplified. * LevelIterator - simplified. * TwoLevelIterator - changed to the new convention. Also fixed a bug that would make SeekForPrev() sometimes silently ignore errors from first_level_iter_. * BlockBasedTableIterator - minor changes. * BlockIter - replaced `SetStatus()` with `Invalidate()` to make sure non-ok BlockIter is always invalid. * PlainTableIterator - some seeks used to not reset status. * CuckooTableIterator - tiny code cleanup. * ManagedIterator - fixed some bugs. * BaseDeltaIterator - changed to the new convention and fixed a bug. * BlobDBIterator - seeks used to not reset status. * KeyConvertingIterator - some small change. Closes https://github.com/facebook/rocksdb/pull/3810 Differential Revision: D7888019 Pulled By: al13n321 fbshipit-source-id: 4aaf6d3421c545d16722a815b2fa2e7912bc851d
2018-05-17 09:44:14 +00:00
if (MaybeFail()) return;
MaybeMutate();
status_ = Status::OK();
// Binary search.
auto it = std::partition_point(
data->entries.begin(), data->entries.end(),
[&](const Entry& e) { return cmp.Compare(e.ikey, target) <= 0; });
iter = (int)(it - data->entries.begin());
--iter;
SkipBackward();
}
void Next() override {
Change and clarify the relationship between Valid(), status() and Seek*() for all iterators. Also fix some bugs Summary: Before this PR, Iterator/InternalIterator may simultaneously have non-ok status() and Valid() = true. That state means that the last operation failed, but the iterator is nevertheless positioned on some unspecified record. Likely intended uses of that are: * If some sst files are corrupted, a normal iterator can be used to read the data from files that are not corrupted. * When using read_tier = kBlockCacheTier, read the data that's in block cache, skipping over the data that is not. However, this behavior wasn't documented well (and until recently the wiki on github had misleading incorrect information). In the code there's a lot of confusion about the relationship between status() and Valid(), and about whether Seek()/SeekToLast()/etc reset the status or not. There were a number of bugs caused by this confusion, both inside rocksdb and in the code that uses rocksdb (including ours). This PR changes the convention to: * If status() is not ok, Valid() always returns false. * Any seek operation resets status. (Before the PR, it depended on iterator type and on particular error.) This does sacrifice the two use cases listed above, but siying said it's ok. Overview of the changes: * A commit that adds missing status checks in MergingIterator. This fixes a bug that actually affects us, and we need it fixed. `DBIteratorTest.NonBlockingIterationBugRepro` explains the scenario. * Changes to lots of iterator types to make all of them conform to the new convention. Some bug fixes along the way. By far the biggest changes are in DBIter, which is a big messy piece of code; I tried to make it less big and messy but mostly failed. * A stress-test for DBIter, to gain some confidence that I didn't break it. It does a few million random operations on the iterator, while occasionally modifying the underlying data (like ForwardIterator does) and occasionally returning non-ok status from internal iterator. To find the iterator types that needed changes I searched for "public .*Iterator" in the code. Here's an overview of all 27 iterator types: Iterators that didn't need changes: * status() is always ok(), or Valid() is always false: MemTableIterator, ModelIter, TestIterator, KVIter (2 classes with this name anonymous namespaces), LoggingForwardVectorIterator, VectorIterator, MockTableIterator, EmptyIterator, EmptyInternalIterator. * Thin wrappers that always pass through Valid() and status(): ArenaWrappedDBIter, TtlIterator, InternalIteratorFromIterator. Iterators with changes (see inline comments for details): * DBIter - an overhaul: - It used to silently skip corrupted keys (`FindParseableKey()`), which seems dangerous. This PR makes it just stop immediately after encountering a corrupted key, just like it would for other kinds of corruption. Let me know if there was actually some deeper meaning in this behavior and I should put it back. - It had a few code paths silently discarding subiterator's status. The stress test caught a few. - The backwards iteration code path was expecting the internal iterator's set of keys to be immutable. It's probably always true in practice at the moment, since ForwardIterator doesn't support backwards iteration, but this PR fixes it anyway. See added DBIteratorTest.ReverseToForwardBug for an example. - Some parts of backwards iteration code path even did things like `assert(iter_->Valid())` after a seek, which is never a safe assumption. - It used to not reset status on seek for some types of errors. - Some simplifications and better comments. - Some things got more complicated from the added error handling. I'm open to ideas for how to make it nicer. * MergingIterator - check status after every operation on every subiterator, and in some places assert that valid subiterators have ok status. * ForwardIterator - changed to the new convention, also slightly simplified. * ForwardLevelIterator - fixed some bugs and simplified. * LevelIterator - simplified. * TwoLevelIterator - changed to the new convention. Also fixed a bug that would make SeekForPrev() sometimes silently ignore errors from first_level_iter_. * BlockBasedTableIterator - minor changes. * BlockIter - replaced `SetStatus()` with `Invalidate()` to make sure non-ok BlockIter is always invalid. * PlainTableIterator - some seeks used to not reset status. * CuckooTableIterator - tiny code cleanup. * ManagedIterator - fixed some bugs. * BaseDeltaIterator - changed to the new convention and fixed a bug. * BlobDBIterator - seeks used to not reset status. * KeyConvertingIterator - some small change. Closes https://github.com/facebook/rocksdb/pull/3810 Differential Revision: D7888019 Pulled By: al13n321 fbshipit-source-id: 4aaf6d3421c545d16722a815b2fa2e7912bc851d
2018-05-17 09:44:14 +00:00
assert(Valid());
if (MaybeFail()) return;
MaybeMutate();
++iter;
SkipForward();
}
void Prev() override {
Change and clarify the relationship between Valid(), status() and Seek*() for all iterators. Also fix some bugs Summary: Before this PR, Iterator/InternalIterator may simultaneously have non-ok status() and Valid() = true. That state means that the last operation failed, but the iterator is nevertheless positioned on some unspecified record. Likely intended uses of that are: * If some sst files are corrupted, a normal iterator can be used to read the data from files that are not corrupted. * When using read_tier = kBlockCacheTier, read the data that's in block cache, skipping over the data that is not. However, this behavior wasn't documented well (and until recently the wiki on github had misleading incorrect information). In the code there's a lot of confusion about the relationship between status() and Valid(), and about whether Seek()/SeekToLast()/etc reset the status or not. There were a number of bugs caused by this confusion, both inside rocksdb and in the code that uses rocksdb (including ours). This PR changes the convention to: * If status() is not ok, Valid() always returns false. * Any seek operation resets status. (Before the PR, it depended on iterator type and on particular error.) This does sacrifice the two use cases listed above, but siying said it's ok. Overview of the changes: * A commit that adds missing status checks in MergingIterator. This fixes a bug that actually affects us, and we need it fixed. `DBIteratorTest.NonBlockingIterationBugRepro` explains the scenario. * Changes to lots of iterator types to make all of them conform to the new convention. Some bug fixes along the way. By far the biggest changes are in DBIter, which is a big messy piece of code; I tried to make it less big and messy but mostly failed. * A stress-test for DBIter, to gain some confidence that I didn't break it. It does a few million random operations on the iterator, while occasionally modifying the underlying data (like ForwardIterator does) and occasionally returning non-ok status from internal iterator. To find the iterator types that needed changes I searched for "public .*Iterator" in the code. Here's an overview of all 27 iterator types: Iterators that didn't need changes: * status() is always ok(), or Valid() is always false: MemTableIterator, ModelIter, TestIterator, KVIter (2 classes with this name anonymous namespaces), LoggingForwardVectorIterator, VectorIterator, MockTableIterator, EmptyIterator, EmptyInternalIterator. * Thin wrappers that always pass through Valid() and status(): ArenaWrappedDBIter, TtlIterator, InternalIteratorFromIterator. Iterators with changes (see inline comments for details): * DBIter - an overhaul: - It used to silently skip corrupted keys (`FindParseableKey()`), which seems dangerous. This PR makes it just stop immediately after encountering a corrupted key, just like it would for other kinds of corruption. Let me know if there was actually some deeper meaning in this behavior and I should put it back. - It had a few code paths silently discarding subiterator's status. The stress test caught a few. - The backwards iteration code path was expecting the internal iterator's set of keys to be immutable. It's probably always true in practice at the moment, since ForwardIterator doesn't support backwards iteration, but this PR fixes it anyway. See added DBIteratorTest.ReverseToForwardBug for an example. - Some parts of backwards iteration code path even did things like `assert(iter_->Valid())` after a seek, which is never a safe assumption. - It used to not reset status on seek for some types of errors. - Some simplifications and better comments. - Some things got more complicated from the added error handling. I'm open to ideas for how to make it nicer. * MergingIterator - check status after every operation on every subiterator, and in some places assert that valid subiterators have ok status. * ForwardIterator - changed to the new convention, also slightly simplified. * ForwardLevelIterator - fixed some bugs and simplified. * LevelIterator - simplified. * TwoLevelIterator - changed to the new convention. Also fixed a bug that would make SeekForPrev() sometimes silently ignore errors from first_level_iter_. * BlockBasedTableIterator - minor changes. * BlockIter - replaced `SetStatus()` with `Invalidate()` to make sure non-ok BlockIter is always invalid. * PlainTableIterator - some seeks used to not reset status. * CuckooTableIterator - tiny code cleanup. * ManagedIterator - fixed some bugs. * BaseDeltaIterator - changed to the new convention and fixed a bug. * BlobDBIterator - seeks used to not reset status. * KeyConvertingIterator - some small change. Closes https://github.com/facebook/rocksdb/pull/3810 Differential Revision: D7888019 Pulled By: al13n321 fbshipit-source-id: 4aaf6d3421c545d16722a815b2fa2e7912bc851d
2018-05-17 09:44:14 +00:00
assert(Valid());
if (MaybeFail()) return;
MaybeMutate();
--iter;
SkipBackward();
}
Slice key() const override {
Change and clarify the relationship between Valid(), status() and Seek*() for all iterators. Also fix some bugs Summary: Before this PR, Iterator/InternalIterator may simultaneously have non-ok status() and Valid() = true. That state means that the last operation failed, but the iterator is nevertheless positioned on some unspecified record. Likely intended uses of that are: * If some sst files are corrupted, a normal iterator can be used to read the data from files that are not corrupted. * When using read_tier = kBlockCacheTier, read the data that's in block cache, skipping over the data that is not. However, this behavior wasn't documented well (and until recently the wiki on github had misleading incorrect information). In the code there's a lot of confusion about the relationship between status() and Valid(), and about whether Seek()/SeekToLast()/etc reset the status or not. There were a number of bugs caused by this confusion, both inside rocksdb and in the code that uses rocksdb (including ours). This PR changes the convention to: * If status() is not ok, Valid() always returns false. * Any seek operation resets status. (Before the PR, it depended on iterator type and on particular error.) This does sacrifice the two use cases listed above, but siying said it's ok. Overview of the changes: * A commit that adds missing status checks in MergingIterator. This fixes a bug that actually affects us, and we need it fixed. `DBIteratorTest.NonBlockingIterationBugRepro` explains the scenario. * Changes to lots of iterator types to make all of them conform to the new convention. Some bug fixes along the way. By far the biggest changes are in DBIter, which is a big messy piece of code; I tried to make it less big and messy but mostly failed. * A stress-test for DBIter, to gain some confidence that I didn't break it. It does a few million random operations on the iterator, while occasionally modifying the underlying data (like ForwardIterator does) and occasionally returning non-ok status from internal iterator. To find the iterator types that needed changes I searched for "public .*Iterator" in the code. Here's an overview of all 27 iterator types: Iterators that didn't need changes: * status() is always ok(), or Valid() is always false: MemTableIterator, ModelIter, TestIterator, KVIter (2 classes with this name anonymous namespaces), LoggingForwardVectorIterator, VectorIterator, MockTableIterator, EmptyIterator, EmptyInternalIterator. * Thin wrappers that always pass through Valid() and status(): ArenaWrappedDBIter, TtlIterator, InternalIteratorFromIterator. Iterators with changes (see inline comments for details): * DBIter - an overhaul: - It used to silently skip corrupted keys (`FindParseableKey()`), which seems dangerous. This PR makes it just stop immediately after encountering a corrupted key, just like it would for other kinds of corruption. Let me know if there was actually some deeper meaning in this behavior and I should put it back. - It had a few code paths silently discarding subiterator's status. The stress test caught a few. - The backwards iteration code path was expecting the internal iterator's set of keys to be immutable. It's probably always true in practice at the moment, since ForwardIterator doesn't support backwards iteration, but this PR fixes it anyway. See added DBIteratorTest.ReverseToForwardBug for an example. - Some parts of backwards iteration code path even did things like `assert(iter_->Valid())` after a seek, which is never a safe assumption. - It used to not reset status on seek for some types of errors. - Some simplifications and better comments. - Some things got more complicated from the added error handling. I'm open to ideas for how to make it nicer. * MergingIterator - check status after every operation on every subiterator, and in some places assert that valid subiterators have ok status. * ForwardIterator - changed to the new convention, also slightly simplified. * ForwardLevelIterator - fixed some bugs and simplified. * LevelIterator - simplified. * TwoLevelIterator - changed to the new convention. Also fixed a bug that would make SeekForPrev() sometimes silently ignore errors from first_level_iter_. * BlockBasedTableIterator - minor changes. * BlockIter - replaced `SetStatus()` with `Invalidate()` to make sure non-ok BlockIter is always invalid. * PlainTableIterator - some seeks used to not reset status. * CuckooTableIterator - tiny code cleanup. * ManagedIterator - fixed some bugs. * BaseDeltaIterator - changed to the new convention and fixed a bug. * BlobDBIterator - seeks used to not reset status. * KeyConvertingIterator - some small change. Closes https://github.com/facebook/rocksdb/pull/3810 Differential Revision: D7888019 Pulled By: al13n321 fbshipit-source-id: 4aaf6d3421c545d16722a815b2fa2e7912bc851d
2018-05-17 09:44:14 +00:00
assert(Valid());
return data->entries[iter].ikey;
}
Slice value() const override {
Change and clarify the relationship between Valid(), status() and Seek*() for all iterators. Also fix some bugs Summary: Before this PR, Iterator/InternalIterator may simultaneously have non-ok status() and Valid() = true. That state means that the last operation failed, but the iterator is nevertheless positioned on some unspecified record. Likely intended uses of that are: * If some sst files are corrupted, a normal iterator can be used to read the data from files that are not corrupted. * When using read_tier = kBlockCacheTier, read the data that's in block cache, skipping over the data that is not. However, this behavior wasn't documented well (and until recently the wiki on github had misleading incorrect information). In the code there's a lot of confusion about the relationship between status() and Valid(), and about whether Seek()/SeekToLast()/etc reset the status or not. There were a number of bugs caused by this confusion, both inside rocksdb and in the code that uses rocksdb (including ours). This PR changes the convention to: * If status() is not ok, Valid() always returns false. * Any seek operation resets status. (Before the PR, it depended on iterator type and on particular error.) This does sacrifice the two use cases listed above, but siying said it's ok. Overview of the changes: * A commit that adds missing status checks in MergingIterator. This fixes a bug that actually affects us, and we need it fixed. `DBIteratorTest.NonBlockingIterationBugRepro` explains the scenario. * Changes to lots of iterator types to make all of them conform to the new convention. Some bug fixes along the way. By far the biggest changes are in DBIter, which is a big messy piece of code; I tried to make it less big and messy but mostly failed. * A stress-test for DBIter, to gain some confidence that I didn't break it. It does a few million random operations on the iterator, while occasionally modifying the underlying data (like ForwardIterator does) and occasionally returning non-ok status from internal iterator. To find the iterator types that needed changes I searched for "public .*Iterator" in the code. Here's an overview of all 27 iterator types: Iterators that didn't need changes: * status() is always ok(), or Valid() is always false: MemTableIterator, ModelIter, TestIterator, KVIter (2 classes with this name anonymous namespaces), LoggingForwardVectorIterator, VectorIterator, MockTableIterator, EmptyIterator, EmptyInternalIterator. * Thin wrappers that always pass through Valid() and status(): ArenaWrappedDBIter, TtlIterator, InternalIteratorFromIterator. Iterators with changes (see inline comments for details): * DBIter - an overhaul: - It used to silently skip corrupted keys (`FindParseableKey()`), which seems dangerous. This PR makes it just stop immediately after encountering a corrupted key, just like it would for other kinds of corruption. Let me know if there was actually some deeper meaning in this behavior and I should put it back. - It had a few code paths silently discarding subiterator's status. The stress test caught a few. - The backwards iteration code path was expecting the internal iterator's set of keys to be immutable. It's probably always true in practice at the moment, since ForwardIterator doesn't support backwards iteration, but this PR fixes it anyway. See added DBIteratorTest.ReverseToForwardBug for an example. - Some parts of backwards iteration code path even did things like `assert(iter_->Valid())` after a seek, which is never a safe assumption. - It used to not reset status on seek for some types of errors. - Some simplifications and better comments. - Some things got more complicated from the added error handling. I'm open to ideas for how to make it nicer. * MergingIterator - check status after every operation on every subiterator, and in some places assert that valid subiterators have ok status. * ForwardIterator - changed to the new convention, also slightly simplified. * ForwardLevelIterator - fixed some bugs and simplified. * LevelIterator - simplified. * TwoLevelIterator - changed to the new convention. Also fixed a bug that would make SeekForPrev() sometimes silently ignore errors from first_level_iter_. * BlockBasedTableIterator - minor changes. * BlockIter - replaced `SetStatus()` with `Invalidate()` to make sure non-ok BlockIter is always invalid. * PlainTableIterator - some seeks used to not reset status. * CuckooTableIterator - tiny code cleanup. * ManagedIterator - fixed some bugs. * BaseDeltaIterator - changed to the new convention and fixed a bug. * BlobDBIterator - seeks used to not reset status. * KeyConvertingIterator - some small change. Closes https://github.com/facebook/rocksdb/pull/3810 Differential Revision: D7888019 Pulled By: al13n321 fbshipit-source-id: 4aaf6d3421c545d16722a815b2fa2e7912bc851d
2018-05-17 09:44:14 +00:00
assert(Valid());
return data->entries[iter].value;
}
bool IsKeyPinned() const override { return true; }
bool IsValuePinned() const override { return true; }
};
// A small reimplementation of DBIter, supporting only some of the features,
// and doing everything in O(log n).
// Skips all keys that are in recently_touched_keys.
struct ReferenceIterator {
Data* data;
uint64_t sequence; // ignore entries with sequence number below this
bool valid = false;
std::string key;
std::string value;
ReferenceIterator(Data* _data, uint64_t _sequence)
: data(_data), sequence(_sequence) {}
bool Valid() const { return valid; }
// Finds the first entry with key
// greater/less/greater-or-equal/less-or-equal than `key`, depending on
// arguments: if `skip`, inequality is strict; if `forward`, it's
// greater/greater-or-equal, otherwise less/less-or-equal.
// Sets `key` to the result.
// If no such key exists, returns false. Doesn't check `visible`.
bool FindNextKey(bool skip, bool forward) {
valid = false;
auto it = std::partition_point(data->entries.begin(), data->entries.end(),
[&](const Entry& e) {
if (forward != skip) {
return e.key < key;
} else {
return e.key <= key;
}
});
if (forward) {
if (it != data->entries.end()) {
key = it->key;
return true;
}
} else {
if (it != data->entries.begin()) {
--it;
key = it->key;
return true;
}
}
return false;
}
bool FindValueForCurrentKey() {
if (data->recently_touched_keys.count(key)) {
return false;
}
// Find the first entry for the key. The caller promises that it exists.
auto it = std::partition_point(data->entries.begin(), data->entries.end(),
[&](const Entry& e) {
if (e.key != key) {
return e.key < key;
}
return e.sequence > sequence;
});
// Find the first visible entry.
for (;; ++it) {
if (it == data->entries.end()) {
return false;
}
Entry& e = *it;
if (e.key != key) {
return false;
}
assert(e.sequence <= sequence);
if (!e.visible) continue;
if (e.type == kTypeDeletion) {
return false;
}
if (e.type == kTypeValue) {
value = e.value;
valid = true;
return true;
}
assert(e.type == kTypeMerge);
break;
}
// Collect merge operands.
std::vector<Slice> operands;
for (; it != data->entries.end(); ++it) {
Entry& e = *it;
if (e.key != key) {
break;
}
assert(e.sequence <= sequence);
if (!e.visible) continue;
if (e.type == kTypeDeletion) {
break;
}
operands.push_back(e.value);
if (e.type == kTypeValue) {
break;
}
}
// Do a merge.
value = operands.back().ToString();
for (int i = (int)operands.size() - 2; i >= 0; --i) {
value.append(",");
value.append(operands[i].data(), operands[i].size());
}
valid = true;
return true;
}
// Start at `key` and move until we encounter a valid value.
// `forward` defines the direction of movement.
// If `skip` is true, we're looking for key not equal to `key`.
void DoTheThing(bool skip, bool forward) {
while (FindNextKey(skip, forward) && !FindValueForCurrentKey()) {
skip = true;
}
}
void Seek(const Slice& target) {
key = target.ToString();
DoTheThing(false, true);
}
void SeekForPrev(const Slice& target) {
key = target.ToString();
DoTheThing(false, false);
}
void SeekToFirst() { Seek(""); }
void SeekToLast() {
key = data->entries.back().key;
DoTheThing(false, false);
}
void Next() {
assert(Valid());
DoTheThing(true, true);
}
void Prev() {
assert(Valid());
DoTheThing(true, false);
}
};
} // namespace
// Use an internal iterator that sometimes returns errors and sometimes
// adds/removes entries on the fly. Do random operations on a DBIter and
// check results.
// TODO: can be improved for more coverage:
// * Override IsKeyPinned() and IsValuePinned() to actually use
// PinnedIteratorManager and check that there's no use-after free.
// * Try different combinations of prefix_extractor, total_order_seek,
// prefix_same_as_start, iterate_lower_bound, iterate_upper_bound.
TEST_F(DBIteratorStressTest, StressTest) {
// We use a deterministic RNG, and everything happens in a single thread.
Random64 rnd(826909345792864532ll);
auto gen_key = [&](int max_key) {
assert(max_key > 0);
Change and clarify the relationship between Valid(), status() and Seek*() for all iterators. Also fix some bugs Summary: Before this PR, Iterator/InternalIterator may simultaneously have non-ok status() and Valid() = true. That state means that the last operation failed, but the iterator is nevertheless positioned on some unspecified record. Likely intended uses of that are: * If some sst files are corrupted, a normal iterator can be used to read the data from files that are not corrupted. * When using read_tier = kBlockCacheTier, read the data that's in block cache, skipping over the data that is not. However, this behavior wasn't documented well (and until recently the wiki on github had misleading incorrect information). In the code there's a lot of confusion about the relationship between status() and Valid(), and about whether Seek()/SeekToLast()/etc reset the status or not. There were a number of bugs caused by this confusion, both inside rocksdb and in the code that uses rocksdb (including ours). This PR changes the convention to: * If status() is not ok, Valid() always returns false. * Any seek operation resets status. (Before the PR, it depended on iterator type and on particular error.) This does sacrifice the two use cases listed above, but siying said it's ok. Overview of the changes: * A commit that adds missing status checks in MergingIterator. This fixes a bug that actually affects us, and we need it fixed. `DBIteratorTest.NonBlockingIterationBugRepro` explains the scenario. * Changes to lots of iterator types to make all of them conform to the new convention. Some bug fixes along the way. By far the biggest changes are in DBIter, which is a big messy piece of code; I tried to make it less big and messy but mostly failed. * A stress-test for DBIter, to gain some confidence that I didn't break it. It does a few million random operations on the iterator, while occasionally modifying the underlying data (like ForwardIterator does) and occasionally returning non-ok status from internal iterator. To find the iterator types that needed changes I searched for "public .*Iterator" in the code. Here's an overview of all 27 iterator types: Iterators that didn't need changes: * status() is always ok(), or Valid() is always false: MemTableIterator, ModelIter, TestIterator, KVIter (2 classes with this name anonymous namespaces), LoggingForwardVectorIterator, VectorIterator, MockTableIterator, EmptyIterator, EmptyInternalIterator. * Thin wrappers that always pass through Valid() and status(): ArenaWrappedDBIter, TtlIterator, InternalIteratorFromIterator. Iterators with changes (see inline comments for details): * DBIter - an overhaul: - It used to silently skip corrupted keys (`FindParseableKey()`), which seems dangerous. This PR makes it just stop immediately after encountering a corrupted key, just like it would for other kinds of corruption. Let me know if there was actually some deeper meaning in this behavior and I should put it back. - It had a few code paths silently discarding subiterator's status. The stress test caught a few. - The backwards iteration code path was expecting the internal iterator's set of keys to be immutable. It's probably always true in practice at the moment, since ForwardIterator doesn't support backwards iteration, but this PR fixes it anyway. See added DBIteratorTest.ReverseToForwardBug for an example. - Some parts of backwards iteration code path even did things like `assert(iter_->Valid())` after a seek, which is never a safe assumption. - It used to not reset status on seek for some types of errors. - Some simplifications and better comments. - Some things got more complicated from the added error handling. I'm open to ideas for how to make it nicer. * MergingIterator - check status after every operation on every subiterator, and in some places assert that valid subiterators have ok status. * ForwardIterator - changed to the new convention, also slightly simplified. * ForwardLevelIterator - fixed some bugs and simplified. * LevelIterator - simplified. * TwoLevelIterator - changed to the new convention. Also fixed a bug that would make SeekForPrev() sometimes silently ignore errors from first_level_iter_. * BlockBasedTableIterator - minor changes. * BlockIter - replaced `SetStatus()` with `Invalidate()` to make sure non-ok BlockIter is always invalid. * PlainTableIterator - some seeks used to not reset status. * CuckooTableIterator - tiny code cleanup. * ManagedIterator - fixed some bugs. * BaseDeltaIterator - changed to the new convention and fixed a bug. * BlobDBIterator - seeks used to not reset status. * KeyConvertingIterator - some small change. Closes https://github.com/facebook/rocksdb/pull/3810 Differential Revision: D7888019 Pulled By: al13n321 fbshipit-source-id: 4aaf6d3421c545d16722a815b2fa2e7912bc851d
2018-05-17 09:44:14 +00:00
int len = 0;
int a = max_key;
while (a) {
a /= 10;
++len;
}
std::string s = ToString(rnd.Next() % static_cast<uint64_t>(max_key));
Change and clarify the relationship between Valid(), status() and Seek*() for all iterators. Also fix some bugs Summary: Before this PR, Iterator/InternalIterator may simultaneously have non-ok status() and Valid() = true. That state means that the last operation failed, but the iterator is nevertheless positioned on some unspecified record. Likely intended uses of that are: * If some sst files are corrupted, a normal iterator can be used to read the data from files that are not corrupted. * When using read_tier = kBlockCacheTier, read the data that's in block cache, skipping over the data that is not. However, this behavior wasn't documented well (and until recently the wiki on github had misleading incorrect information). In the code there's a lot of confusion about the relationship between status() and Valid(), and about whether Seek()/SeekToLast()/etc reset the status or not. There were a number of bugs caused by this confusion, both inside rocksdb and in the code that uses rocksdb (including ours). This PR changes the convention to: * If status() is not ok, Valid() always returns false. * Any seek operation resets status. (Before the PR, it depended on iterator type and on particular error.) This does sacrifice the two use cases listed above, but siying said it's ok. Overview of the changes: * A commit that adds missing status checks in MergingIterator. This fixes a bug that actually affects us, and we need it fixed. `DBIteratorTest.NonBlockingIterationBugRepro` explains the scenario. * Changes to lots of iterator types to make all of them conform to the new convention. Some bug fixes along the way. By far the biggest changes are in DBIter, which is a big messy piece of code; I tried to make it less big and messy but mostly failed. * A stress-test for DBIter, to gain some confidence that I didn't break it. It does a few million random operations on the iterator, while occasionally modifying the underlying data (like ForwardIterator does) and occasionally returning non-ok status from internal iterator. To find the iterator types that needed changes I searched for "public .*Iterator" in the code. Here's an overview of all 27 iterator types: Iterators that didn't need changes: * status() is always ok(), or Valid() is always false: MemTableIterator, ModelIter, TestIterator, KVIter (2 classes with this name anonymous namespaces), LoggingForwardVectorIterator, VectorIterator, MockTableIterator, EmptyIterator, EmptyInternalIterator. * Thin wrappers that always pass through Valid() and status(): ArenaWrappedDBIter, TtlIterator, InternalIteratorFromIterator. Iterators with changes (see inline comments for details): * DBIter - an overhaul: - It used to silently skip corrupted keys (`FindParseableKey()`), which seems dangerous. This PR makes it just stop immediately after encountering a corrupted key, just like it would for other kinds of corruption. Let me know if there was actually some deeper meaning in this behavior and I should put it back. - It had a few code paths silently discarding subiterator's status. The stress test caught a few. - The backwards iteration code path was expecting the internal iterator's set of keys to be immutable. It's probably always true in practice at the moment, since ForwardIterator doesn't support backwards iteration, but this PR fixes it anyway. See added DBIteratorTest.ReverseToForwardBug for an example. - Some parts of backwards iteration code path even did things like `assert(iter_->Valid())` after a seek, which is never a safe assumption. - It used to not reset status on seek for some types of errors. - Some simplifications and better comments. - Some things got more complicated from the added error handling. I'm open to ideas for how to make it nicer. * MergingIterator - check status after every operation on every subiterator, and in some places assert that valid subiterators have ok status. * ForwardIterator - changed to the new convention, also slightly simplified. * ForwardLevelIterator - fixed some bugs and simplified. * LevelIterator - simplified. * TwoLevelIterator - changed to the new convention. Also fixed a bug that would make SeekForPrev() sometimes silently ignore errors from first_level_iter_. * BlockBasedTableIterator - minor changes. * BlockIter - replaced `SetStatus()` with `Invalidate()` to make sure non-ok BlockIter is always invalid. * PlainTableIterator - some seeks used to not reset status. * CuckooTableIterator - tiny code cleanup. * ManagedIterator - fixed some bugs. * BaseDeltaIterator - changed to the new convention and fixed a bug. * BlobDBIterator - seeks used to not reset status. * KeyConvertingIterator - some small change. Closes https://github.com/facebook/rocksdb/pull/3810 Differential Revision: D7888019 Pulled By: al13n321 fbshipit-source-id: 4aaf6d3421c545d16722a815b2fa2e7912bc851d
2018-05-17 09:44:14 +00:00
s.insert(0, len - (int)s.size(), '0');
return s;
};
Options options;
options.merge_operator = MergeOperators::CreateFromStringId("stringappend");
ReadOptions ropt;
size_t num_matching = 0;
size_t num_at_end = 0;
size_t num_not_ok = 0;
size_t num_recently_removed = 0;
// Number of iterations for each combination of parameters
// (there are ~250 of those).
// Tweak this to change the test run time.
// As of the time of writing, the test takes ~4 seconds for value of 5000.
const int num_iterations = 5000;
// Enable this to print all the operations for debugging.
bool trace = FLAGS_verbose;
for (int num_entries : {5, 10, 100}) {
for (double key_space : {0.1, 1.0, 3.0}) {
for (ValueType prevalent_entry_type :
{kTypeValue, kTypeDeletion, kTypeMerge}) {
for (double error_probability : {0.01, 0.1}) {
for (double mutation_probability : {0.01, 0.5}) {
for (double target_hidden_fraction : {0.1, 0.5}) {
std::string trace_str =
"entries: " + ToString(num_entries) +
", key_space: " + ToString(key_space) +
", error_probability: " + ToString(error_probability) +
", mutation_probability: " + ToString(mutation_probability) +
", target_hidden_fraction: " +
ToString(target_hidden_fraction);
SCOPED_TRACE(trace_str);
if (trace) {
std::cout << trace_str << std::endl;
}
// Generate data.
Data data;
int max_key = (int)(num_entries * key_space) + 1;
for (int i = 0; i < num_entries; ++i) {
Entry e;
e.key = gen_key(max_key);
if (rnd.Next() % 10 != 0) {
e.type = prevalent_entry_type;
} else {
const ValueType types[] = {kTypeValue, kTypeDeletion,
kTypeMerge};
e.type =
types[rnd.Next() % (sizeof(types) / sizeof(types[0]))];
}
e.sequence = i;
e.value = "v" + ToString(i);
ParsedInternalKey internal_key(e.key, e.sequence, e.type);
AppendInternalKey(&e.ikey, internal_key);
data.entries.push_back(e);
}
std::sort(data.entries.begin(), data.entries.end());
if (trace) {
std::cout << "entries:";
for (size_t i = 0; i < data.entries.size(); ++i) {
Entry& e = data.entries[i];
std::cout
<< "\n idx " << i << ": \"" << e.key << "\": \""
<< e.value << "\" seq: " << e.sequence << " type: "
<< (e.type == kTypeValue
? "val"
: e.type == kTypeDeletion ? "del" : "merge");
}
std::cout << std::endl;
}
std::unique_ptr<Iterator> db_iter;
std::unique_ptr<ReferenceIterator> ref_iter;
for (int iteration = 0; iteration < num_iterations; ++iteration) {
SCOPED_TRACE(iteration);
// Create a new iterator every ~30 operations.
if (db_iter == nullptr || rnd.Next() % 30 == 0) {
uint64_t sequence = rnd.Next() % (data.entries.size() + 2);
ref_iter.reset(new ReferenceIterator(&data, sequence));
if (trace) {
std::cout << "new iterator, seq: " << sequence << std::endl;
}
auto internal_iter =
new StressTestIterator(&data, &rnd, BytewiseComparator());
internal_iter->error_probability = error_probability;
internal_iter->mutation_probability = mutation_probability;
internal_iter->target_hidden_fraction =
target_hidden_fraction;
internal_iter->trace = trace;
db_iter.reset(NewDBIterator(
env_, ropt, ImmutableCFOptions(options),
MutableCFOptions(options), BytewiseComparator(),
internal_iter, sequence,
Change and clarify the relationship between Valid(), status() and Seek*() for all iterators. Also fix some bugs Summary: Before this PR, Iterator/InternalIterator may simultaneously have non-ok status() and Valid() = true. That state means that the last operation failed, but the iterator is nevertheless positioned on some unspecified record. Likely intended uses of that are: * If some sst files are corrupted, a normal iterator can be used to read the data from files that are not corrupted. * When using read_tier = kBlockCacheTier, read the data that's in block cache, skipping over the data that is not. However, this behavior wasn't documented well (and until recently the wiki on github had misleading incorrect information). In the code there's a lot of confusion about the relationship between status() and Valid(), and about whether Seek()/SeekToLast()/etc reset the status or not. There were a number of bugs caused by this confusion, both inside rocksdb and in the code that uses rocksdb (including ours). This PR changes the convention to: * If status() is not ok, Valid() always returns false. * Any seek operation resets status. (Before the PR, it depended on iterator type and on particular error.) This does sacrifice the two use cases listed above, but siying said it's ok. Overview of the changes: * A commit that adds missing status checks in MergingIterator. This fixes a bug that actually affects us, and we need it fixed. `DBIteratorTest.NonBlockingIterationBugRepro` explains the scenario. * Changes to lots of iterator types to make all of them conform to the new convention. Some bug fixes along the way. By far the biggest changes are in DBIter, which is a big messy piece of code; I tried to make it less big and messy but mostly failed. * A stress-test for DBIter, to gain some confidence that I didn't break it. It does a few million random operations on the iterator, while occasionally modifying the underlying data (like ForwardIterator does) and occasionally returning non-ok status from internal iterator. To find the iterator types that needed changes I searched for "public .*Iterator" in the code. Here's an overview of all 27 iterator types: Iterators that didn't need changes: * status() is always ok(), or Valid() is always false: MemTableIterator, ModelIter, TestIterator, KVIter (2 classes with this name anonymous namespaces), LoggingForwardVectorIterator, VectorIterator, MockTableIterator, EmptyIterator, EmptyInternalIterator. * Thin wrappers that always pass through Valid() and status(): ArenaWrappedDBIter, TtlIterator, InternalIteratorFromIterator. Iterators with changes (see inline comments for details): * DBIter - an overhaul: - It used to silently skip corrupted keys (`FindParseableKey()`), which seems dangerous. This PR makes it just stop immediately after encountering a corrupted key, just like it would for other kinds of corruption. Let me know if there was actually some deeper meaning in this behavior and I should put it back. - It had a few code paths silently discarding subiterator's status. The stress test caught a few. - The backwards iteration code path was expecting the internal iterator's set of keys to be immutable. It's probably always true in practice at the moment, since ForwardIterator doesn't support backwards iteration, but this PR fixes it anyway. See added DBIteratorTest.ReverseToForwardBug for an example. - Some parts of backwards iteration code path even did things like `assert(iter_->Valid())` after a seek, which is never a safe assumption. - It used to not reset status on seek for some types of errors. - Some simplifications and better comments. - Some things got more complicated from the added error handling. I'm open to ideas for how to make it nicer. * MergingIterator - check status after every operation on every subiterator, and in some places assert that valid subiterators have ok status. * ForwardIterator - changed to the new convention, also slightly simplified. * ForwardLevelIterator - fixed some bugs and simplified. * LevelIterator - simplified. * TwoLevelIterator - changed to the new convention. Also fixed a bug that would make SeekForPrev() sometimes silently ignore errors from first_level_iter_. * BlockBasedTableIterator - minor changes. * BlockIter - replaced `SetStatus()` with `Invalidate()` to make sure non-ok BlockIter is always invalid. * PlainTableIterator - some seeks used to not reset status. * CuckooTableIterator - tiny code cleanup. * ManagedIterator - fixed some bugs. * BaseDeltaIterator - changed to the new convention and fixed a bug. * BlobDBIterator - seeks used to not reset status. * KeyConvertingIterator - some small change. Closes https://github.com/facebook/rocksdb/pull/3810 Differential Revision: D7888019 Pulled By: al13n321 fbshipit-source-id: 4aaf6d3421c545d16722a815b2fa2e7912bc851d
2018-05-17 09:44:14 +00:00
options.max_sequential_skip_in_iterations,
nullptr /*read_callback*/));
}
// Do a random operation. It's important to do it on ref_it
// later than on db_iter to make sure ref_it sees the correct
// recently_touched_keys.
std::string old_key;
bool forward = rnd.Next() % 2 > 0;
// Do Next()/Prev() ~90% of the time.
bool seek = !ref_iter->Valid() || rnd.Next() % 10 == 0;
if (trace) {
std::cout << iteration << ": ";
}
if (!seek) {
assert(db_iter->Valid());
old_key = ref_iter->key;
if (trace) {
std::cout << (forward ? "Next" : "Prev") << std::endl;
}
if (forward) {
db_iter->Next();
ref_iter->Next();
} else {
db_iter->Prev();
ref_iter->Prev();
}
} else {
data.recently_touched_keys.clear();
// Do SeekToFirst less often than Seek.
if (rnd.Next() % 4 == 0) {
if (trace) {
std::cout << (forward ? "SeekToFirst" : "SeekToLast")
<< std::endl;
}
if (forward) {
old_key = "";
db_iter->SeekToFirst();
ref_iter->SeekToFirst();
} else {
old_key = data.entries.back().key;
db_iter->SeekToLast();
ref_iter->SeekToLast();
}
} else {
old_key = gen_key(max_key);
if (trace) {
std::cout << (forward ? "Seek" : "SeekForPrev") << " \""
<< old_key << '"' << std::endl;
}
if (forward) {
db_iter->Seek(old_key);
ref_iter->Seek(old_key);
} else {
db_iter->SeekForPrev(old_key);
ref_iter->SeekForPrev(old_key);
}
}
}
// Check the result.
if (db_iter->Valid()) {
ASSERT_TRUE(db_iter->status().ok());
if (data.recently_touched_keys.count(
db_iter->key().ToString())) {
// Ended on a key that may have been mutated during the
// operation. Reference iterator skips such keys, so we
// can't check the exact result.
// Check that the key moved in the right direction.
if (forward) {
if (seek)
ASSERT_GE(db_iter->key().ToString(), old_key);
else
ASSERT_GT(db_iter->key().ToString(), old_key);
} else {
if (seek)
ASSERT_LE(db_iter->key().ToString(), old_key);
else
ASSERT_LT(db_iter->key().ToString(), old_key);
}
if (ref_iter->Valid()) {
// Check that DBIter didn't miss any non-mutated key.
if (forward) {
ASSERT_LT(db_iter->key().ToString(), ref_iter->key);
} else {
ASSERT_GT(db_iter->key().ToString(), ref_iter->key);
}
}
// Tell the next iteration of the loop to reseek the
// iterators.
ref_iter->valid = false;
++num_recently_removed;
} else {
ASSERT_TRUE(ref_iter->Valid());
ASSERT_EQ(ref_iter->key, db_iter->key().ToString());
ASSERT_EQ(ref_iter->value, db_iter->value());
++num_matching;
}
} else if (db_iter->status().ok()) {
ASSERT_FALSE(ref_iter->Valid());
++num_at_end;
} else {
// Non-ok status. Nothing to check here.
// Tell the next iteration of the loop to reseek the
// iterators.
ref_iter->valid = false;
++num_not_ok;
}
}
}
}
}
}
}
}
// Check that all cases were hit many times.
EXPECT_GT(num_matching, 10000);
EXPECT_GT(num_at_end, 10000);
EXPECT_GT(num_not_ok, 10000);
EXPECT_GT(num_recently_removed, 10000);
std::cout << "stats:\n exact matches: " << num_matching
<< "\n end reached: " << num_at_end
<< "\n non-ok status: " << num_not_ok
<< "\n mutated on the fly: " << num_recently_removed << std::endl;
}
} // namespace rocksdb
int main(int argc, char** argv) {
::testing::InitGoogleTest(&argc, argv);
ParseCommandLineFlags(&argc, &argv, true);
return RUN_ALL_TESTS();
}