rocksdb/cache/cache_key.cc
Peter Dillinger 86a1e3e0e7 Derive cache keys from SST unique IDs (#10394)
Summary:
... so that cache keys can be derived from DB manifest data
before reading the file from storage--so that every part of the file
can potentially go in a persistent cache.

See updated comments in cache_key.cc for technical details. Importantly,
the new cache key encoding uses some fancy but efficient math to pack
data into the cache key without depending on the sizes of the various
pieces. This simplifies some existing code creating cache keys, like
cache warming before the file size is known.

This should provide us an essentially permanent mapping between SST
unique IDs and base cache keys, with the ability to "upgrade" SST
unique IDs (and thus cache keys) with new SST format_versions.

These cache keys are of similar, perhaps indistinguishable quality to
the previous generation. Before this change (see "corrected" days
between collision):

```
./cache_bench -stress_cache_key -sck_keep_bits=43
18 collisions after 2 x 90 days, est 10 days between (1.15292e+19 corrected)
```

After this change (keep 43 bits, up through 50, to validate "trajectory"
is ok on "corrected" days between collision):
```
19 collisions after 3 x 90 days, est 14.2105 days between (1.63836e+19 corrected)
16 collisions after 5 x 90 days, est 28.125 days between (1.6213e+19 corrected)
15 collisions after 7 x 90 days, est 42 days between (1.21057e+19 corrected)
15 collisions after 17 x 90 days, est 102 days between (1.46997e+19 corrected)
15 collisions after 49 x 90 days, est 294 days between (2.11849e+19 corrected)
15 collisions after 62 x 90 days, est 372 days between (1.34027e+19 corrected)
15 collisions after 53 x 90 days, est 318 days between (5.72858e+18 corrected)
15 collisions after 309 x 90 days, est 1854 days between (1.66994e+19 corrected)
```

However, the change does modify (probably weaken) the "guaranteed unique" promise from this

> SST files generated in a single process are guaranteed to have unique cache keys, unless/until number session ids * max file number = 2**86

to this (see https://github.com/facebook/rocksdb/issues/10388)

> With the DB id limitation, we only have nice guaranteed unique cache keys for files generated in a single process until biggest session_id_counter and offset_in_file reach combined 64 bits

I don't think this is a practical concern, though.

Pull Request resolved: https://github.com/facebook/rocksdb/pull/10394

Test Plan: unit tests updated, see simulation results above

Reviewed By: jay-zhuang

Differential Revision: D38667529

Pulled By: pdillinger

fbshipit-source-id: 49af3fe7f47e5b61162809a78b76c769fd519fba
2022-08-12 13:49:49 -07:00

365 lines
18 KiB
C++

// Copyright (c) Facebook, Inc. and its affiliates. 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 "cache/cache_key.h"
#include <algorithm>
#include <atomic>
#include "rocksdb/cache.h"
#include "table/unique_id_impl.h"
#include "util/hash.h"
#include "util/math.h"
namespace ROCKSDB_NAMESPACE {
// Value space plan for CacheKey:
//
// file_num_etc64_ | offset_etc64_ | Only generated by
// ---------------+---------------+------------------------------------------
// 0 | 0 | Reserved for "empty" CacheKey()
// 0 | > 0, < 1<<63 | CreateUniqueForCacheLifetime
// 0 | >= 1<<63 | CreateUniqueForProcessLifetime
// > 0 | any | OffsetableCacheKey.WithOffset
CacheKey CacheKey::CreateUniqueForCacheLifetime(Cache *cache) {
// +1 so that we can reserve all zeros for "unset" cache key
uint64_t id = cache->NewId() + 1;
// Ensure we don't collide with CreateUniqueForProcessLifetime
assert((id >> 63) == 0U);
return CacheKey(0, id);
}
CacheKey CacheKey::CreateUniqueForProcessLifetime() {
// To avoid colliding with CreateUniqueForCacheLifetime, assuming
// Cache::NewId counts up from zero, here we count down from UINT64_MAX.
// If this ever becomes a point of contention, we could sub-divide the
// space and use CoreLocalArray.
static std::atomic<uint64_t> counter{UINT64_MAX};
uint64_t id = counter.fetch_sub(1, std::memory_order_relaxed);
// Ensure we don't collide with CreateUniqueForCacheLifetime
assert((id >> 63) == 1U);
return CacheKey(0, id);
}
// How we generate CacheKeys and base OffsetableCacheKey, assuming that
// db_session_ids are generated from a base_session_id and
// session_id_counter (by SemiStructuredUniqueIdGen+EncodeSessionId
// in DBImpl::GenerateDbSessionId):
//
// Conceptual inputs:
// db_id (unstructured, from GenerateRawUniqueId or equiv)
// * could be shared between cloned DBs but rare
// * could be constant, if session id suffices
// base_session_id (unstructured, from GenerateRawUniqueId)
// session_id_counter (structured)
// * usually much smaller than 2**24
// orig_file_number (structured)
// * usually smaller than 2**24
// offset_in_file (structured, might skip lots of values)
// * usually smaller than 2**32
//
// Overall approach (see https://github.com/pdillinger/unique_id for
// background):
//
// First, we have three "structured" values, up to 64 bits each, that we
// need to fit, without losses, into 128 bits. In practice, the values will
// be small enough that they should fit. For example, applications generating
// large SST files (large offsets) will naturally produce fewer files (small
// file numbers). But we don't know ahead of time what bounds the values will
// have.
//
// Second, we have unstructured inputs that enable distinct RocksDB processes
// to pick a random point in space, likely very different from others. Xoring
// the structured with the unstructured give us a cache key that is
// structurally distinct between related keys (e.g. same file or same RocksDB
// process) and distinct with high probability between unrelated keys.
//
// The problem of packing three structured values into the space for two is
// complicated by the fact that we want to derive cache keys from SST unique
// IDs, which have already combined structured and unstructured inputs in a
// practically inseparable way. And we want a base cache key that works
// with an offset of any size. So basically, we need to encode these three
// structured values, each up to 64 bits, into 128 bits without knowing any
// of their sizes. The DownwardInvolution() function gives us a mechanism to
// accomplish this. (See its properties in math.h.) Specifically, for inputs
// a, b, and c:
// lower64 = DownwardInvolution(a) ^ ReverseBits(b);
// upper64 = c ^ ReverseBits(a);
// The 128-bit output is unique assuming there exist some i, j, and k
// where a < 2**i, b < 2**j, c < 2**k, i <= 64, j <= 64, k <= 64, and
// i + j + k <= 128. In other words, as long as there exist some bounds
// that would allow us to pack the bits of a, b, and c into the output
// if we know the bound, we can generate unique outputs without knowing
// those bounds. To validate this claim, the inversion function (given
// the bounds) has been implemented in CacheKeyDecoder in
// db_block_cache_test.cc.
//
// With that in mind, the outputs in terms of the conceptual inputs look
// like this, using bitwise-xor of the constituent pieces, low bits on left:
//
// |------------------------- file_num_etc64 -------------------------|
// | +++++++++ base_session_id (lower 64 bits, involution) +++++++++ |
// |-----------------------------------------------------------------|
// | session_id_counter (involution) ..... | |
// |-----------------------------------------------------------------|
// | hash of: ++++++++++++++++++++++++++++++++++++++++++++++++++++++ |
// | * base_session_id (upper ~39 bits) |
// | * db_id (~122 bits entropy) |
// |-----------------------------------------------------------------|
// | | ..... orig_file_number (reversed) |
// |-----------------------------------------------------------------|
//
//
// |------------------------- offset_etc64 --------------------------|
// | ++++++++++ base_session_id (lower 64 bits, reversed) ++++++++++ |
// |-----------------------------------------------------------------|
// | | ..... session_id_counter (reversed) |
// |-----------------------------------------------------------------|
// | offset_in_file ............... | |
// |-----------------------------------------------------------------|
//
// Some oddities or inconveniences of this layout are due to deriving
// the "base" cache key (without offset) from the SST unique ID (see
// GetSstInternalUniqueId). Specifically,
// * Lower 64 of base_session_id occurs in both output words (ok but
// weird)
// * The inclusion of db_id is bad for the conditions under which we
// can guarantee uniqueness, but could be useful in some cases with
// few small files per process, to make up for db session id only having
// ~103 bits of entropy.
//
// In fact, if DB ids were not involved, we would be guaranteed unique
// cache keys for files generated in a single process until total bits for
// biggest session_id_counter, orig_file_number, and offset_in_file
// reach 128 bits.
//
// With the DB id limitation, we only have nice guaranteed unique cache
// keys for files generated in a single process until biggest
// session_id_counter and offset_in_file reach combined 64 bits. This
// is quite good in practice because we can have millions of DB Opens
// with terabyte size SST files, or billions of DB Opens with gigabyte
// size SST files.
//
// One of the considerations in the translation between existing SST unique
// IDs and base cache keys is supporting better SST unique IDs in a future
// format_version. If we use a process-wide file counter instead of
// session counter and file numbers, we only need to combine two 64-bit values
// instead of three. But we don't want to track unique ID versions in the
// manifest, so we want to keep the same translation layer between SST unique
// IDs and base cache keys, even with updated SST unique IDs. If the new
// unique IDs put the file counter where the orig_file_number was, and
// use no structured field where session_id_counter was, then our translation
// layer works fine for two structured fields as well as three (for
// compatibility). The small computation for the translation (one
// DownwardInvolution(), two ReverseBits(), both ~log(64) instructions deep)
// is negligible for computing as part of SST file reader open.
//
// More on how https://github.com/pdillinger/unique_id applies here:
// Every bit of output always includes "unstructured" uniqueness bits and
// often combines with "structured" uniqueness bits. The "unstructured" bits
// change infrequently: only when we cannot guarantee our state tracking for
// "structured" uniqueness hasn't been cloned. Using a static
// SemiStructuredUniqueIdGen for db_session_ids, this means we only get an
// "all new" session id when a new process uses RocksDB. (Between processes,
// we don't know if a DB or other persistent storage has been cloned. We
// assume that if VM hot cloning is used, subsequently generated SST files
// do not interact.) Within a process, only the session_lower of the
// db_session_id changes incrementally ("structured" uniqueness).
//
// This basically means that our offsets, counters and file numbers allow us
// to do somewhat "better than random" (birthday paradox) while in the
// degenerate case of completely new session for each tiny file, we still
// have strong uniqueness properties from the birthday paradox, with ~103
// bit session IDs or up to 128 bits entropy with different DB IDs sharing a
// cache.
//
// More collision probability analysis:
// Suppose a RocksDB host generates (generously) 2 GB/s (10TB data, 17 DWPD)
// with average process/session lifetime of (pessimistically) 4 minutes.
// In 180 days (generous allowable data lifespan), we generate 31 million GB
// of data, or 2^55 bytes, and 2^16 "all new" session IDs.
//
// First, suppose this is in a single DB (lifetime 180 days):
// 128 bits cache key size
// - 55 <- ideal size for byte offsets + file numbers
// - 2 <- bits for offsets and file numbers not exactly powers of two
// + 2 <- bits saved not using byte offsets in BlockBasedTable::GetCacheKey
// ----
// 73 <- bits remaining for distinguishing session IDs
// The probability of a collision in 73 bits of session ID data is less than
// 1 in 2**(73 - (2 * 16)), or roughly 1 in a trillion. And this assumes all
// data from the last 180 days is in cache for potential collision, and that
// cache keys under each session id exhaustively cover the remaining 57 bits
// while in reality they'll only cover a small fraction of it.
//
// Although data could be transferred between hosts, each host has its own
// cache and we are already assuming a high rate of "all new" session ids.
// So this doesn't really change the collision calculation. Across a fleet
// of 1 million, each with <1 in a trillion collision possibility,
// fleetwide collision probability is <1 in a million.
//
// Now suppose we have many DBs per host, say 2**10, with same host-wide write
// rate and process/session lifetime. File numbers will be ~10 bits smaller
// and we will have 2**10 times as many session IDs because of simultaneous
// lifetimes. So now collision chance is less than 1 in 2**(83 - (2 * 26)),
// or roughly 1 in a billion.
//
// Suppose instead we generated random or hashed cache keys for each
// (compressed) block. For 1KB compressed block size, that is 2^45 cache keys
// in 180 days. Collision probability is more easily estimated at roughly
// 1 in 2**(128 - (2 * 45)) or roughly 1 in a trillion (assuming all
// data from the last 180 days is in cache, but NOT the other assumption
// for the 1 in a trillion estimate above).
//
//
// Collision probability estimation through simulation:
// A tool ./cache_bench -stress_cache_key broadly simulates host-wide cache
// activity over many months, by making some pessimistic simplifying
// assumptions. See class StressCacheKey in cache_bench_tool.cc for details.
// Here is some sample output with
// `./cache_bench -stress_cache_key -sck_keep_bits=43`:
//
// Total cache or DBs size: 32TiB Writing 925.926 MiB/s or 76.2939TiB/day
// Multiply by 1.15292e+18 to correct for simulation losses (but still
// assume whole file cached)
//
// These come from default settings of 2.5M files per day of 32 MB each, and
// `-sck_keep_bits=43` means that to represent a single file, we are only
// keeping 43 bits of the 128-bit (base) cache key. With file size of 2**25
// contiguous keys (pessimistic), our simulation is about 2\*\*(128-43-25) or
// about 1 billion billion times more prone to collision than reality.
//
// More default assumptions, relatively pessimistic:
// * 100 DBs in same process (doesn't matter much)
// * Re-open DB in same process (new session ID related to old session ID) on
// average every 100 files generated
// * Restart process (all new session IDs unrelated to old) 24 times per day
//
// After enough data, we get a result at the end (-sck_keep_bits=43):
//
// (keep 43 bits) 18 collisions after 2 x 90 days, est 10 days between
// (1.15292e+19 corrected)
//
// If we believe the (pessimistic) simulation and the mathematical
// extrapolation, we would need to run a billion machines all for 11 billion
// days to expect a cache key collision. To help verify that our extrapolation
// ("corrected") is robust, we can make our simulation more precise by
// increasing the "keep" bits, which takes more running time to get enough
// collision data:
//
// (keep 44 bits) 16 collisions after 5 x 90 days, est 28.125 days between
// (1.6213e+19 corrected)
// (keep 45 bits) 15 collisions after 7 x 90 days, est 42 days between
// (1.21057e+19 corrected)
// (keep 46 bits) 15 collisions after 17 x 90 days, est 102 days between
// (1.46997e+19 corrected)
// (keep 47 bits) 15 collisions after 49 x 90 days, est 294 days between
// (2.11849e+19 corrected)
//
// The extrapolated prediction seems to be within noise (sampling error).
//
// With the `-sck_randomize` option, we can see that typical workloads like
// above have lower collision probability than "random" cache keys (note:
// offsets still non-randomized) by a modest amount (roughly 2-3x less
// collision prone than random), which should make us reasonably comfortable
// even in "degenerate" cases (e.g. repeatedly launch a process to generate
// one file with SstFileWriter):
//
// (rand 43 bits) 22 collisions after 1 x 90 days, est 4.09091 days between
// (4.7165e+18 corrected)
//
// We can see that with more frequent process restarts,
// -sck_restarts_per_day=5000, which means more all-new session IDs, we get
// closer to the "random" cache key performance:
//
// 15 collisions after 1 x 90 days, est 6 days between (6.91753e+18 corrected)
//
// And with less frequent process restarts and re-opens,
// -sck_restarts_per_day=1 -sck_reopen_nfiles=1000, we get lower collision
// probability:
//
// 18 collisions after 8 x 90 days, est 40 days between (4.61169e+19 corrected)
//
// Other tests have been run to validate other conditions behave as expected,
// never behaving "worse than random" unless we start chopping off structured
// data.
//
// Conclusion: Even in extreme cases, rapidly burning through "all new" IDs
// that only arise when a new process is started, the chance of any cache key
// collisions in a giant fleet of machines is negligible. Especially when
// processes live for hours or days, the chance of a cache key collision is
// likely more plausibly due to bad hardware than to bad luck in random
// session ID data. Software defects are surely more likely to cause corruption
// than both of those.
//
// TODO: Nevertheless / regardless, an efficient way to detect (and thus
// quantify) block cache corruptions, including collisions, should be added.
OffsetableCacheKey::OffsetableCacheKey(const std::string &db_id,
const std::string &db_session_id,
uint64_t file_number) {
UniqueId64x2 internal_id;
Status s = GetSstInternalUniqueId(db_id, db_session_id, file_number,
&internal_id, /*force=*/true);
assert(s.ok());
*this = FromInternalUniqueId(&internal_id);
}
OffsetableCacheKey OffsetableCacheKey::FromInternalUniqueId(UniqueIdPtr id) {
uint64_t session_lower = id.ptr[0];
uint64_t file_num_etc = id.ptr[1];
#ifndef NDEBUG
bool is_empty = session_lower == 0 && file_num_etc == 0;
#endif
// Although DBImpl guarantees (in recent versions) that session_lower is not
// zero, that's not entirely sufficient to guarantee that file_num_etc64_ is
// not zero (so that the 0 case can be used by CacheKey::CreateUnique*)
// However, if we are given an "empty" id as input, then we should produce
// "empty" as output.
// As a consequence, this function is only bijective assuming
// id[0] == 0 only if id[1] == 0.
if (session_lower == 0U) {
session_lower = file_num_etc;
}
// See comments above for how DownwardInvolution and ReverseBits
// make this function invertible under various assumptions.
OffsetableCacheKey rv;
rv.file_num_etc64_ =
DownwardInvolution(session_lower) ^ ReverseBits(file_num_etc);
rv.offset_etc64_ = ReverseBits(session_lower);
// Because of these transformations and needing to allow arbitrary
// offset (thus, second 64 bits of cache key might be 0), we need to
// make some correction to ensure the first 64 bits is not 0.
// Fortunately, the transformation ensures the second 64 bits is not 0
// for non-empty base key, so we can swap in the case one is 0 without
// breaking bijectivity (assuming condition above).
assert(is_empty || rv.offset_etc64_ > 0);
if (rv.file_num_etc64_ == 0) {
std::swap(rv.file_num_etc64_, rv.offset_etc64_);
}
assert(is_empty || rv.file_num_etc64_ > 0);
return rv;
}
// Inverse of FromInternalUniqueId (assuming file_num_etc64 == 0 only if
// offset_etc64 == 0)
UniqueId64x2 OffsetableCacheKey::ToInternalUniqueId() {
uint64_t a = file_num_etc64_;
uint64_t b = offset_etc64_;
if (b == 0) {
std::swap(a, b);
}
UniqueId64x2 rv;
rv[0] = ReverseBits(b);
rv[1] = ReverseBits(a ^ DownwardInvolution(rv[0]));
return rv;
}
} // namespace ROCKSDB_NAMESPACE