gotosocial/vendor/codeberg.org/gruf/go-structr/cache.go
kim 84279f6a6a
[performance] cache more database calls, reduce required database calls overall (#3290)
* improvements to caching for lists and relationship to accounts / follows

* fix nil panic in AddToList()

* ensure list related caches are correctly invalidated

* ensure returned ID lists are ordered correctly

* bump go-structr to v0.8.9 (returns early if zero uncached keys to be loaded)

* remove zero checks in uncached key load functions (go-structr now handles this)

* fix issues after rebase on upstream/main

* update the expected return order of CSV exports (since list entries are now down by entry creation date)

* rename some funcs, allow deleting list entries for multiple follow IDs at a time, fix up more tests

* use returning statements on delete to get cache invalidation info

* fixes to recent database delete changes

* fix broken list entries delete sql

* remove unused db function

* update remainder of delete functions to behave in similar way, some other small tweaks

* fix delete user sql, allow returning on err no entries

* uncomment + fix list database tests

* update remaining list tests

* update envparsing test

* add comments to each specific key being invalidated

* add more cache invalidation explanatory comments

* whoops; actually delete poll votes from database in the DeletePollByID() func

* remove added but-commented-out field

* improved comment regarding paging being disabled

* make cache invalidation comments match what's actually happening

* fix up delete query comments to match what is happening

* rename function to read a bit better

* don't use ErrNoEntries on delete when not needed (it's only needed for a RETURNING call)

* update function name in test

* move list exclusivity check to AFTER eligibility check. use log.Panic() instead of panic()

* use the poll_id column in poll_votes for selecting votes in poll ID

* fix function name
2024-09-16 16:46:09 +00:00

678 lines
14 KiB
Go

package structr
import (
"context"
"errors"
"reflect"
"sync"
"unsafe"
)
// DefaultIgnoreErr is the default function used to
// ignore (i.e. not cache) incoming error results during
// Load() calls. By default ignores context pkg errors.
func DefaultIgnoreErr(err error) bool {
return errors.Is(err, context.Canceled) ||
errors.Is(err, context.DeadlineExceeded)
}
// CacheConfig defines config vars
// for initializing a struct cache.
type CacheConfig[StructType any] struct {
// IgnoreErr defines which errors to
// ignore (i.e. not cache) returned
// from load function callback calls.
// This may be left as nil, on which
// DefaultIgnoreErr will be used.
IgnoreErr func(error) bool
// Copy provides a means of copying
// cached values, to ensure returned values
// do not share memory with those in cache.
Copy func(StructType) StructType
// Invalidate is called when cache values
// (NOT errors) are invalidated, either
// as the values passed to Put() / Store(),
// or by the keys by calls to Invalidate().
Invalidate func(StructType)
// Indices defines indices to create
// in the Cache for the receiving
// generic struct type parameter.
Indices []IndexConfig
// MaxSize defines the maximum number
// of items allowed in the Cache at
// one time, before old items start
// getting evicted.
MaxSize int
}
// Cache provides a structure cache with automated
// indexing and lookups by any initialization-defined
// combination of fields. This also supports caching
// of negative results (errors!) returned by LoadOne().
type Cache[StructType any] struct {
// hook functions.
ignore func(error) bool
copy func(StructType) StructType
invalid func(StructType)
// keeps track of all indexed items,
// in order of last recently used (LRU).
lru list
// indices used in storing passed struct
// types by user defined sets of fields.
indices []Index
// max cache size, imposes size
// limit on the lruList in order
// to evict old entries.
maxSize int
// protective mutex, guards:
// - Cache{}.lruList
// - Index{}.data
// - Cache{} hook fns
mutex sync.Mutex
}
// Init initializes the cache with given configuration
// including struct fields to index, and necessary fns.
func (c *Cache[T]) Init(config CacheConfig[T]) {
t := reflect.TypeOf((*T)(nil)).Elem()
if len(config.Indices) == 0 {
panic("no indices provided")
}
if config.IgnoreErr == nil {
config.IgnoreErr = DefaultIgnoreErr
}
if config.Copy == nil {
panic("copy function must be provided")
}
if config.MaxSize < 2 {
panic("minimum cache size is 2 for LRU to work")
}
// Safely copy over
// provided config.
c.mutex.Lock()
c.indices = make([]Index, len(config.Indices))
for i, cfg := range config.Indices {
c.indices[i].ptr = unsafe.Pointer(c)
c.indices[i].init(t, cfg, config.MaxSize)
}
c.ignore = config.IgnoreErr
c.copy = config.Copy
c.invalid = config.Invalidate
c.maxSize = config.MaxSize
c.mutex.Unlock()
}
// Index selects index with given name from cache, else panics.
func (c *Cache[T]) Index(name string) *Index {
for i := range c.indices {
if c.indices[i].name == name {
return &c.indices[i]
}
}
panic("unknown index: " + name)
}
// GetOne fetches value from cache stored under index, using precalculated index key.
func (c *Cache[T]) GetOne(index *Index, key Key) (T, bool) {
values := c.Get(index, key)
if len(values) == 0 {
var zero T
return zero, false
}
return values[0], true
}
// Get fetches values from the cache stored under index, using precalculated index keys.
func (c *Cache[T]) Get(index *Index, keys ...Key) []T {
if index == nil {
panic("no index given")
} else if index.ptr != unsafe.Pointer(c) {
panic("invalid index for cache")
}
// Preallocate expected ret slice.
values := make([]T, 0, len(keys))
// Acquire lock.
c.mutex.Lock()
defer c.mutex.Unlock()
// Check cache init.
if c.copy == nil {
panic("not initialized")
}
for i := range keys {
// Concatenate all *values* from cached items.
index.get(keys[i].key, func(item *indexed_item) {
if value, ok := item.data.(T); ok {
// Append value COPY.
value = c.copy(value)
values = append(values, value)
// Push to front of LRU list, USING
// THE ITEM'S LRU ENTRY, NOT THE
// INDEX KEY ENTRY. VERY IMPORTANT!!
c.lru.move_front(&item.elem)
}
})
}
return values
}
// Put will insert the given values into cache,
// calling any invalidate hook on each value.
func (c *Cache[T]) Put(values ...T) {
// Acquire lock.
c.mutex.Lock()
// Wrap unlock to only do once.
unlock := once(c.mutex.Unlock)
defer unlock()
// Check cache init.
if c.copy == nil {
panic("not initialized")
}
// Store all passed values.
for i := range values {
c.store_value(
nil, "",
values[i],
)
}
// Get func ptrs.
invalid := c.invalid
// Done with
// the lock.
unlock()
if invalid != nil {
// Pass all invalidated values
// to given user hook (if set).
for _, value := range values {
invalid(value)
}
}
}
// LoadOneBy fetches one result from the cache stored under index, using precalculated index key.
// In the case that no result is found, provided load callback will be used to hydrate the cache.
func (c *Cache[T]) LoadOne(index *Index, key Key, load func() (T, error)) (T, error) {
if index == nil {
panic("no index given")
} else if index.ptr != unsafe.Pointer(c) {
panic("invalid index for cache")
} else if !is_unique(index.flags) {
panic("cannot get one by non-unique index")
}
var (
// whether an item was found
// (and so val / err are set).
ok bool
// separate value / error ptrs
// as the item is liable to
// change outside of lock.
val T
err error
)
// Acquire lock.
c.mutex.Lock()
// Wrap unlock to only do once.
unlock := once(c.mutex.Unlock)
defer unlock()
// Check init'd.
if c.copy == nil ||
c.ignore == nil {
panic("not initialized")
}
// Get item indexed at key.
item := index.get_one(key)
if ok = (item != nil); ok {
var is bool
if val, is = item.data.(T); is {
// Set value COPY.
val = c.copy(val)
// Push to front of LRU list, USING
// THE ITEM'S LRU ENTRY, NOT THE
// INDEX KEY ENTRY. VERY IMPORTANT!!
c.lru.move_front(&item.elem)
} else {
// Attempt to return error.
err, _ = item.data.(error)
}
}
// Get func ptrs.
ignore := c.ignore
// Done with
// the lock.
unlock()
if ok {
// item found!
return val, err
}
// Load new result.
val, err = load()
// Check for ignored error types.
if err != nil && ignore(err) {
return val, err
}
// Acquire lock.
c.mutex.Lock()
// Index this new loaded item.
// Note this handles copying of
// the provided value, so it is
// safe for us to return as-is.
if err != nil {
c.store_error(index, key.key, err)
} else {
c.store_value(index, key.key, val)
}
// Done with lock.
c.mutex.Unlock()
return val, err
}
// Load fetches values from the cache stored under index, using precalculated index keys. The cache will attempt to
// results with values stored under keys, passing keys with uncached results to the provider load callback to further
// hydrate the cache with missing results. Cached error results not included or returned by this function.
func (c *Cache[T]) Load(index *Index, keys []Key, load func([]Key) ([]T, error)) ([]T, error) {
if index == nil {
panic("no index given")
} else if index.ptr != unsafe.Pointer(c) {
panic("invalid index for cache")
}
// Preallocate expected ret slice.
values := make([]T, 0, len(keys))
// Acquire lock.
c.mutex.Lock()
// Wrap unlock to only do once.
unlock := once(c.mutex.Unlock)
defer unlock()
// Check init'd.
if c.copy == nil {
panic("not initialized")
}
for i := 0; i < len(keys); {
// Value length before
// any below appends.
before := len(values)
// Concatenate all *values* from cached items.
index.get(keys[i].key, func(item *indexed_item) {
if value, ok := item.data.(T); ok {
// Append value COPY.
value = c.copy(value)
values = append(values, value)
// Push to front of LRU list, USING
// THE ITEM'S LRU ENTRY, NOT THE
// INDEX KEY ENTRY. VERY IMPORTANT!!
c.lru.move_front(&item.elem)
}
})
// Only if values changed did
// we actually find anything.
if len(values) != before {
// We found values at key,
// drop key from the slice.
copy(keys[i:], keys[i+1:])
keys = keys[:len(keys)-1]
continue
}
// Iter
i++
}
// Done with
// the lock.
unlock()
if len(keys) == 0 {
// We loaded everything!
return values, nil
}
// Load uncached values.
uncached, err := load(keys)
if err != nil {
return nil, err
}
// Acquire lock.
c.mutex.Lock()
// Store all uncached values.
for i := range uncached {
c.store_value(
nil, "",
uncached[i],
)
}
// Done with lock.
c.mutex.Unlock()
// Append uncached to return values.
values = append(values, uncached...)
return values, nil
}
// Store will call the given store callback, on non-error then
// passing the provided value to the Put() function. On error
// return the value is still passed to stored invalidate hook.
func (c *Cache[T]) Store(value T, store func() error) error {
// Store value.
err := store()
if err != nil {
// Get func ptrs.
c.mutex.Lock()
invalid := c.invalid
c.mutex.Unlock()
// On error don't store
// value, but still pass
// to invalidate hook.
if invalid != nil {
invalid(value)
}
return err
}
// Store value.
c.Put(value)
return nil
}
// Invalidate invalidates all results stored under index keys.
func (c *Cache[T]) Invalidate(index *Index, keys ...Key) {
if index == nil {
panic("no index given")
} else if index.ptr != unsafe.Pointer(c) {
panic("invalid index for cache")
}
// Acquire lock.
c.mutex.Lock()
// Preallocate expected ret slice.
values := make([]T, 0, len(keys))
for i := range keys {
// Delete all items under key from index, collecting
// value items and dropping them from all their indices.
index.delete(keys[i].key, func(item *indexed_item) {
if value, ok := item.data.(T); ok {
// No need to copy, as item
// being deleted from cache.
values = append(values, value)
}
// Delete cached.
c.delete(item)
})
}
// Get func ptrs.
invalid := c.invalid
// Done with lock.
c.mutex.Unlock()
if invalid != nil {
// Pass all invalidated values
// to given user hook (if set).
for _, value := range values {
invalid(value)
}
}
}
// Trim will truncate the cache to ensure it
// stays within given percentage of MaxSize.
func (c *Cache[T]) Trim(perc float64) {
// Acquire lock.
c.mutex.Lock()
// Calculate number of cache items to drop.
max := (perc / 100) * float64(c.maxSize)
diff := c.lru.len - int(max)
if diff <= 0 {
// Trim not needed.
c.mutex.Unlock()
return
}
// Iterate over 'diff' items
// from back (oldest) of cache.
for i := 0; i < diff; i++ {
// Get oldest LRU elem.
oldest := c.lru.tail
if oldest == nil {
// reached
// end.
break
}
// Drop oldest item from cache.
item := (*indexed_item)(oldest.data)
c.delete(item)
}
// Compact index data stores.
for i := range c.indices {
c.indices[i].data.Compact()
}
// Done with lock.
c.mutex.Unlock()
}
// Clear empties the cache by calling .Trim(0).
func (c *Cache[T]) Clear() { c.Trim(0) }
// Len returns the current length of cache.
func (c *Cache[T]) Len() int {
c.mutex.Lock()
l := c.lru.len
c.mutex.Unlock()
return l
}
// Debug returns debug stats about cache.
func (c *Cache[T]) Debug() map[string]any {
m := make(map[string]any)
c.mutex.Lock()
m["lru"] = c.lru.len
indices := make(map[string]any)
m["indices"] = indices
for i := range c.indices {
var n uint64
for _, l := range c.indices[i].data.m {
n += uint64(l.len)
}
indices[c.indices[i].name] = n
}
c.mutex.Unlock()
return m
}
// Cap returns the maximum capacity (size) of cache.
func (c *Cache[T]) Cap() int {
c.mutex.Lock()
m := c.maxSize
c.mutex.Unlock()
return m
}
func (c *Cache[T]) store_value(index *Index, key string, value T) {
// Alloc new index item.
item := new_indexed_item()
if cap(item.indexed) < len(c.indices) {
// Preallocate item indices slice to prevent Go auto
// allocating overlying large slices we don't need.
item.indexed = make([]*index_entry, 0, len(c.indices))
}
// Create COPY of value.
value = c.copy(value)
item.data = value
if index != nil {
// Append item to index.
index.append(key, item)
}
// Get ptr to value data.
ptr := unsafe.Pointer(&value)
// Acquire key buf.
buf := new_buffer()
for i := range c.indices {
// Get current index ptr.
idx := &(c.indices[i])
if idx == index {
// Already stored under
// this index, ignore.
continue
}
// Extract fields comprising index key.
parts := extract_fields(ptr, idx.fields)
if parts == nil {
continue
}
// Calculate index key.
key := idx.key(buf, parts)
if key == "" {
continue
}
// Append item to index.
idx.append(key, item)
}
// Add item to main lru list.
c.lru.push_front(&item.elem)
// Done with buf.
free_buffer(buf)
if c.lru.len > c.maxSize {
// Cache has hit max size!
// Drop the oldest element.
ptr := c.lru.tail.data
item := (*indexed_item)(ptr)
c.delete(item)
}
}
func (c *Cache[T]) store_error(index *Index, key string, err error) {
if index == nil {
// nothing we
// can do here.
return
}
// Alloc new index item.
item := new_indexed_item()
if cap(item.indexed) < len(c.indices) {
// Preallocate item indices slice to prevent Go auto
// allocating overlying large slices we don't need.
item.indexed = make([]*index_entry, 0, len(c.indices))
}
// Set error val.
item.data = err
// Append item to index.
index.append(key, item)
// Add item to main lru list.
c.lru.push_front(&item.elem)
if c.lru.len > c.maxSize {
// Cache has hit max size!
// Drop the oldest element.
ptr := c.lru.tail.data
item := (*indexed_item)(ptr)
c.delete(item)
}
}
func (c *Cache[T]) delete(item *indexed_item) {
for len(item.indexed) != 0 {
// Pop last indexed entry from list.
entry := item.indexed[len(item.indexed)-1]
item.indexed = item.indexed[:len(item.indexed)-1]
// Drop index_entry from index.
entry.index.delete_entry(entry)
}
// Drop entry from lru list.
c.lru.remove(&item.elem)
// Free now-unused item.
free_indexed_item(item)
}