We thought stack traces were mildly expensive. We were very wrong. Due to a complicated issue in the Go runtime, simultaneous requests for stack traces end up contending on a lock even when they're not actually contending on any resources. I've filed a ticket in the Go issue tracker for this: https://github.com/golang/go/issues/56400 In the mean time: Under some workloads, we were seeing 85% of all CPU time go into the stack backtraces, of which 81% went into the contention on those locks. But even if you take away the contention, that leaves us with 4/19 of all CPU time in our code going into building those stack backtraces. That's a lot of overhead for a feature we virtually never use. We might consider adding a backtrace functionality here, possibly using `runtime.Callers` which is much lower overhead, and allows us to generate a backtrace on demand (no argument values available, but then, we never read those because they're unformatted hex values), but I don't think it's actually very informative to know what the stack traces were of the Tx; they don't necessarily reflect the current state of any ongoing use of the Tx, so we can't necessarily correlate them to goroutine stack dumps, and so on. |
||
|---|---|---|
| .. | ||
| cfg | ||
| rbf/testdata/check/bad-freelist | ||
| testdata/check | ||
| array.go | ||
| cursor.go | ||
| cursor_internal_test.go | ||
| cursor_test.go | ||
| cursorx.go | ||
| db.go | ||
| db_test.go | ||
| dot.go | ||
| helpers_test.go | ||
| ingest_test.go | ||
| page_map.go | ||
| rbf.go | ||
| rbf_test.go | ||
| README.md | ||
| tx.go | ||
| tx_test.go | ||
| util.go | ||
| util_test.go | ||
Roaring B-tree Format
The RBF format represents a Roaring bitmap whose containers are stored in the leafs of a b-tree. This allows the bitmap to be efficiently queried & updated.
File Format
The RBF file is divided into equal 8KB pages. Each page after the meta page is numbered incrementally from 1 to 2^31.
Pages can be one of the following types:
- Meta page: contains header information.
- Branch page: contains pointers to lower branch & leaf pages.
- Leaf page: contains array and RLE container data.
- Bitmap page: contains bitmap container data.
All integer values are little endian encoded.
Page header
Every page type except the bitmap page contains the following header:
[4] page number
[4] flags (indicates the type of the page)
Meta page
The meta page contains the following header:
[4] magic (\xFFRBF)
[4] flags
[4] page count
[8] wal ID
[4] root records pgno
[4] freelist pgno
Root Records page
A list of all b-tree names & their respective root page numbers are stored in root record pages. Once a bitmap root is created, it is never moved so the root record pages only need to be rewritten when creating, renaming, or deleting a b-tree. If records exceed the size of a page then they are overflowed to additional pages.
[4] page number
[4] flags
[4] overflow pgno
[*] bitmap records
Each bitmap record is represented as:
[4] pgno
[2] name size
[*] name
All bitmap records are loaded into memory when the file is opened.
Branch page
The branch page contains the following header:
[4] page number
[4] flags
[2] cell count
[*] cell index (2 * cell count)
[*] padding for 4-byte alignment
Each cell is formatted as:
[8] highbits
[4] flags
[4] page number
Leaf page
The leaf page contains the following header:
[4] page number
[4] flags
[2] cell count
[*] cell index (2 * cell count)
The leaf page contains a series of cells with the header of:
[8] highbits
[4] flag
[4] child count
[*] array or RLE data or Handle (a pageno) to Bitmap Data
Bitmap Data page
The data for the bitmap data page takes up the entire 8KB.
Proof of Concept Notes
The following are notes made that are temporary for the RBF format. This will change as development progresses:
- Transaction support is deferred
- WAL support is deferred
Pattern of branch splits when adding data in ascending sorted order.
In this example, fan-out is restricted to 2 to make the drawings easy and the splits obvious. The data leaves are only allowed one roaring.Container key (ckey) in this example.
Each frame adds the next datum: A,B,C,D,E,... in order.
The letter represent data leaves, while numbers represent branch pages. The one exception is the first frame where the root is a leaf with data. Every frame after has normal branch at the root.
NB: there are only three places pages get written on addition: a) writeRoot b) putLeafCell c) putBranchCells
add A: root 3 A
add B: root 3 4 5 A B
add C: this sequence of updates occurs
-
putLeafCell writes leaf B to page 5
-
putLeafCell writes leaf C to page 6
-
putBranchCells writes branch page 7 with children 4,5
-
putBranchCells writes branch page 8 with child 6
-
writeRoot writes branch cells to pgno 3, children: 7,8
root 3 7 8 4 5 6 A B C
add D:
root
3
7 8
4 5 6 9
A B C D