[performance] rewrite timelines to rely on new timeline cache type (#3941)

* start work rewriting timeline cache type

* further work rewriting timeline caching

* more work integration new timeline code

* remove old code

* add local timeline, fix up merge conflicts

* remove old use of go-bytes

* implement new timeline code into more areas of codebase, pull in latest go-mangler, go-mutexes, go-structr

* remove old timeline package, add local timeline cache

* remove references to old timeline types that needed starting up in tests

* start adding page validation

* fix test-identified timeline cache package issues

* fix up more tests, fix missing required changes, etc

* add exclusion for test.out in gitignore

* clarify some things better in code comments

* tweak cache size limits

* fix list timeline cache fetching

* further list timeline fixes

* linter, ssssssssshhhhhhhhhhhh please

* fix linter hints

* reslice the output if it's beyond length of 'lim'

* remove old timeline initialization code, bump go-structr to v0.9.4

* continued from previous commit

* improved code comments

* don't allow multiple entries for BoostOfID values to prevent repeated boosts of same boosts

* finish writing more code comments

* some variable renaming, for ease of following

* change the way we update lo,hi paging values during timeline load

* improved code comments for updated / returned lo , hi paging values

* finish writing code comments for the StatusTimeline{} type itself

* fill in more code comments

* update go-structr version to latest with changed timeline unique indexing logic

* have a local and public timeline *per user*

* rewrite calls to public / local timeline calls

* remove the zero length check, as lo, hi values might still be set

* simplify timeline cache loading, fix lo/hi returns, fix timeline invalidation side-effects missing for some federated actions

* swap the lo, hi values 🤦

* add (now) missing slice reverse of tag timeline statuses when paging ASC

* remove local / public caches (is out of scope for this work), share more timeline code

* remove unnecessary change

* again, remove more unused code

* remove unused function to appease the linter

* move boost checking to prepare function

* fix use of timeline.lastOrder, fix incorrect range functions used

* remove comments for repeat code

* remove the boost logic from prepare function

* do a maximum of 5 loads, not 10

* add repeat boost filtering logic, update go-structr, general improvements

* more code comments

* add important note

* fix timeline tests now that timelines are returned in page order

* remove unused field

* add StatusTimeline{} tests

* add more status timeline tests

* start adding preloading support

* ensure repeat boosts are marked in preloaded entries

* share a bunch of the database load code in timeline cache, don't clear timelines on relationship change

* add logic to allow dynamic clear / preloading of timelines

* comment-out unused functions, but leave in place as we might end-up using them

* fix timeline preload state check

* much improved status timeline code comments

* more code comments, don't bother inserting statuses if timeline not preloaded

* shift around some logic to make sure things aren't accidentally left set

* finish writing code comments

* remove trim-after-insert behaviour

* fix-up some comments referring to old logic

* remove unsetting of lo, hi

* fix preload repeatBoost checking logic

* don't return on status filter errors, these are usually transient

* better concurrency safety in Clear() and Done()

* fix test broken due to addition of preloader

* fix repeatBoost logic that doesn't account for already-hidden repeatBoosts

* ensure edit submodels are dropped on cache insertion

* update code-comment to expand CAS accronym

* use a plus1hULID() instead of 24h

* remove unused functions

* add note that public / local timeline requester can be nil

* fix incorrect visibility filtering of tag timeline statuses

* ensure we filter home timeline statuses on local only

* some small re-orderings to confirm query params in correct places

* fix the local only home timeline filter func
This commit is contained in:
kim
2025-04-26 09:56:15 +00:00
committed by GitHub
parent ffde1b150f
commit 6a6a499333
108 changed files with 2935 additions and 5213 deletions

View File

@@ -1,6 +1,7 @@
package structr
import (
"fmt"
"os"
"reflect"
"strings"
@@ -222,10 +223,10 @@ func (i *Index) get(key string, hook func(*indexed_item)) {
func (i *Index) key(buf *byteutil.Buffer, parts []unsafe.Pointer) string {
buf.B = buf.B[:0]
if len(parts) != len(i.fields) {
panicf("incorrect number key parts: want=%d received=%d",
panic(fmt.Sprintf("incorrect number key parts: want=%d received=%d",
len(i.fields),
len(parts),
)
))
}
if !allow_zero(i.flags) {
for x, field := range i.fields {

View File

@@ -70,7 +70,7 @@ func find_field(t reflect.Type, names []string) (sfield struct_field) {
name := names[0]
names = names[1:]
if !is_exported(name) {
panicf("field is not exported: %s", name)
panic(fmt.Sprintf("field is not exported: %s", name))
}
return name
}
@@ -94,7 +94,7 @@ func find_field(t reflect.Type, names []string) (sfield struct_field) {
// Check for valid struct type.
if t.Kind() != reflect.Struct {
panicf("field %s is not struct (or ptr-to): %s", t, name)
panic(fmt.Sprintf("field %s is not struct (or ptr-to): %s", t, name))
}
var ok bool
@@ -102,7 +102,7 @@ func find_field(t reflect.Type, names []string) (sfield struct_field) {
// Look for next field by name.
field, ok = t.FieldByName(name)
if !ok {
panicf("unknown field: %s", name)
panic(fmt.Sprintf("unknown field: %s", name))
}
// Set next offset value.
@@ -258,11 +258,6 @@ func eface_data(a any) unsafe.Pointer {
return (*eface)(unsafe.Pointer(&a)).data
}
// panicf provides a panic with string formatting.
func panicf(format string, args ...any) {
panic(fmt.Sprintf(format, args...))
}
// assert can be called to indicated a block
// of code should not be able to be reached,
// it returns a BUG report with callsite.

View File

@@ -190,7 +190,8 @@ func (t *Timeline[T, PK]) Select(min, max *PK, length *int, dir Direction) (valu
// Insert will insert the given values into the timeline,
// calling any set invalidate hook on each inserted value.
func (t *Timeline[T, PK]) Insert(values ...T) {
// Returns current list length after performing inserts.
func (t *Timeline[T, PK]) Insert(values ...T) int {
// Acquire lock.
t.mutex.Lock()
@@ -269,6 +270,10 @@ func (t *Timeline[T, PK]) Insert(values ...T) {
// Get func ptrs.
invalid := t.invalid
// Get length AFTER
// insert to return.
len := t.list.len
// Done with lock.
t.mutex.Unlock()
@@ -279,6 +284,8 @@ func (t *Timeline[T, PK]) Insert(values ...T) {
invalid(value)
}
}
return len
}
// Invalidate invalidates all entries stored in index under given keys.
@@ -336,8 +343,8 @@ func (t *Timeline[T, PK]) Invalidate(index *Index, keys ...Key) {
//
// Please note that the entire Timeline{} will be locked for the duration of the range
// operation, i.e. from the beginning of the first yield call until the end of the last.
func (t *Timeline[T, PK]) Range(dir Direction) func(yield func(T) bool) {
return func(yield func(T) bool) {
func (t *Timeline[T, PK]) Range(dir Direction) func(yield func(index int, value T) bool) {
return func(yield func(int, T) bool) {
if t.copy == nil {
panic("not initialized")
} else if yield == nil {
@@ -348,7 +355,9 @@ func (t *Timeline[T, PK]) Range(dir Direction) func(yield func(T) bool) {
t.mutex.Lock()
defer t.mutex.Unlock()
var i int
switch dir {
case Asc:
// Iterate through linked list from bottom (i.e. tail).
for prev := t.list.tail; prev != nil; prev = prev.prev {
@@ -360,9 +369,12 @@ func (t *Timeline[T, PK]) Range(dir Direction) func(yield func(T) bool) {
value := t.copy(item.data.(T))
// Pass to given function.
if !yield(value) {
if !yield(i, value) {
break
}
// Iter
i++
}
case Desc:
@@ -376,9 +388,12 @@ func (t *Timeline[T, PK]) Range(dir Direction) func(yield func(T) bool) {
value := t.copy(item.data.(T))
// Pass to given function.
if !yield(value) {
if !yield(i, value) {
break
}
// Iter
i++
}
}
}
@@ -390,8 +405,8 @@ func (t *Timeline[T, PK]) Range(dir Direction) func(yield func(T) bool) {
//
// Please note that the entire Timeline{} will be locked for the duration of the range
// operation, i.e. from the beginning of the first yield call until the end of the last.
func (t *Timeline[T, PK]) RangeUnsafe(dir Direction) func(yield func(T) bool) {
return func(yield func(T) bool) {
func (t *Timeline[T, PK]) RangeUnsafe(dir Direction) func(yield func(index int, value T) bool) {
return func(yield func(int, T) bool) {
if t.copy == nil {
panic("not initialized")
} else if yield == nil {
@@ -402,7 +417,9 @@ func (t *Timeline[T, PK]) RangeUnsafe(dir Direction) func(yield func(T) bool) {
t.mutex.Lock()
defer t.mutex.Unlock()
var i int
switch dir {
case Asc:
// Iterate through linked list from bottom (i.e. tail).
for prev := t.list.tail; prev != nil; prev = prev.prev {
@@ -411,9 +428,12 @@ func (t *Timeline[T, PK]) RangeUnsafe(dir Direction) func(yield func(T) bool) {
item := (*timeline_item)(prev.data)
// Pass to given function.
if !yield(item.data.(T)) {
if !yield(i, item.data.(T)) {
break
}
// Iter
i++
}
case Desc:
@@ -424,9 +444,12 @@ func (t *Timeline[T, PK]) RangeUnsafe(dir Direction) func(yield func(T) bool) {
item := (*timeline_item)(next.data)
// Pass to given function.
if !yield(item.data.(T)) {
if !yield(i, item.data.(T)) {
break
}
// Iter
i++
}
}
}
@@ -1033,6 +1056,9 @@ indexing:
// checking for collisions.
if !idx.add(key, i_item) {
// This key already appears
// in this unique index. So
// drop new timeline item.
t.delete(t_item)
free_buffer(buf)
return last