-
Notifications
You must be signed in to change notification settings - Fork 2
feat(sqs): HT-FIFO partitioned-keyspace constructors (Phase 3.D PR 3) #703
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Merged
Merged
Changes from all commits
Commits
Show all changes
2 commits
Select commit
Hold shift + click to select a range
File filter
Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
There are no files selected for viewing
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
|
|
@@ -49,6 +49,52 @@ const ( | |
| SqsQueueTombstonePrefix = "!sqs|queue|tombstone|" | ||
| ) | ||
|
|
||
| // HT-FIFO partitioned-keyspace discriminator. Per the §3.1 design in | ||
| // docs/design/2026_04_26_proposed_sqs_split_queue_fifo.md, partitioned | ||
| // FIFO queues live in a separate keyspace so the legacy single- | ||
| // partition layout can stay byte-identical on disk: | ||
| // | ||
| // legacy: !sqs|msg|<family>|<queue>|<gen>|<rest> | ||
| // partitioned: !sqs|msg|<family>|p|<queue>|<partition>|<gen>|<rest> | ||
| // | ||
| // The literal "p|" segment is the discriminator. validateQueueName | ||
| // rejects "|" in queue names, so a legacy "!sqs|msg|data|<queue>|..." | ||
| // can never collide with a partitioned "!sqs|msg|data|p|<queue>|..." | ||
| // — the queue-name segment is base64-raw-URL-encoded (see | ||
| // encodeSQSSegment) and cannot start with the literal ASCII byte 'p' | ||
| // followed by '|'. | ||
| // | ||
| // Each partitioned constructor terminates the variable-length | ||
| // queue-name segment with a '|' before the fixed-width partition | ||
| // uint32. Without that delimiter, a prefix scan for queue "q" would | ||
| // also match queue "q1" because base64("q") is a strict byte prefix | ||
| // of base64("q1"). The discriminator inserts the '|' into the prefix | ||
| // itself; the per-constructor terminator inserts it after the queue. | ||
| const sqsPartitionedDiscriminator = "p|" | ||
|
|
||
| // sqsPartitionedQueueTerminator is appended after the encoded queue | ||
| // name in every partitioned key. It mirrors the role the fixed-width | ||
| // generation suffix plays in tombstone keys: a hard end-of-segment | ||
| // marker that prevents queue-name prefix collisions during scans. | ||
| // '|' is safe by construction — validateQueueName rejects raw '|', | ||
| // and base64.RawURLEncoding never emits '|' (it uses A-Z, a-z, 0-9, | ||
| // '-', '_'). | ||
| const sqsPartitionedQueueTerminator = '|' | ||
|
|
||
| // SqsPartitionedMsg*Prefix mirrors each legacy SqsMsg*Prefix with the | ||
| // partitioned-keyspace discriminator inserted. Defined as full string | ||
| // constants (rather than runtime concatenation in each constructor) | ||
| // so the byte-layout invariant is asserted by the type system: a | ||
| // future rename of the discriminator must touch the constants here, | ||
| // not 6+ scattered string concatenations. | ||
| const ( | ||
| SqsPartitionedMsgDataPrefix = "!sqs|msg|data|" + sqsPartitionedDiscriminator | ||
| SqsPartitionedMsgVisPrefix = "!sqs|msg|vis|" + sqsPartitionedDiscriminator | ||
| SqsPartitionedMsgDedupPrefix = "!sqs|msg|dedup|" + sqsPartitionedDiscriminator | ||
| SqsPartitionedMsgGroupPrefix = "!sqs|msg|group|" + sqsPartitionedDiscriminator | ||
| SqsPartitionedMsgByAgePrefix = "!sqs|msg|byage|" + sqsPartitionedDiscriminator | ||
| ) | ||
|
|
||
| func sqsQueueMetaKey(queueName string) []byte { | ||
| return []byte(SqsQueueMetaPrefix + encodeSQSSegment(queueName)) | ||
| } | ||
|
|
@@ -209,3 +255,194 @@ func queueNameFromMetaKey(key []byte) (string, bool) { | |
| } | ||
| return name, true | ||
| } | ||
|
|
||
| // ---------------------- HT-FIFO partitioned keyspace ---------------------- | ||
| // | ||
| // The constructors below mirror the legacy sqsMsg*Key family with a | ||
| // partition uint32 inserted between the queue segment and the | ||
| // generation. The legacy keyspace is unchanged on disk, so existing | ||
| // queues and Standard queues stay byte-identical — these helpers are | ||
| // reachable only when meta.PartitionCount > 1, and the §11 PR 2 | ||
| // dormancy gate currently rejects that at CreateQueue. The data plane | ||
| // dispatch lands together with the gate-lift in PR 5. | ||
| // | ||
| // Each helper appends the partition as a fixed-width big-endian | ||
| // uint32 so prefix scans `!sqs|msg|<family>|p|<queue>|<partition>|` | ||
| // can pick exactly one partition's keys without touching its | ||
| // neighbours. | ||
|
|
||
| // sqsPartitionedMsgDataKey builds the data-record key for a | ||
| // partitioned FIFO queue. | ||
| func sqsPartitionedMsgDataKey(queueName string, partition uint32, gen uint64, messageID string) []byte { | ||
| buf := make([]byte, 0, len(SqsPartitionedMsgDataPrefix)+sqsKeyCapLarge) | ||
| buf = append(buf, SqsPartitionedMsgDataPrefix...) | ||
| buf = append(buf, encodeSQSSegment(queueName)...) | ||
| buf = append(buf, sqsPartitionedQueueTerminator) | ||
| buf = appendU32(buf, partition) | ||
| buf = appendU64(buf, gen) | ||
| buf = append(buf, encodeSQSSegment(messageID)...) | ||
| return buf | ||
| } | ||
|
|
||
| // sqsPartitionedMsgVisKey builds the visibility-index key for a | ||
| // partitioned FIFO queue. | ||
| func sqsPartitionedMsgVisKey(queueName string, partition uint32, gen uint64, visibleAtMillis int64, messageID string) []byte { | ||
| buf := make([]byte, 0, len(SqsPartitionedMsgVisPrefix)+sqsKeyCapLarge) | ||
| buf = append(buf, SqsPartitionedMsgVisPrefix...) | ||
| buf = append(buf, encodeSQSSegment(queueName)...) | ||
| buf = append(buf, sqsPartitionedQueueTerminator) | ||
| buf = appendU32(buf, partition) | ||
| buf = appendU64(buf, gen) | ||
| buf = appendU64(buf, uint64MaxZero(visibleAtMillis)) | ||
| buf = append(buf, encodeSQSSegment(messageID)...) | ||
| return buf | ||
| } | ||
|
|
||
| // sqsPartitionedMsgVisPrefixForQueue returns the prefix of every | ||
| // vis-index key for a single (queue, partition, gen) triple. The | ||
| // partition fan-out scans this prefix on each partition independently. | ||
| func sqsPartitionedMsgVisPrefixForQueue(queueName string, partition uint32, gen uint64) []byte { | ||
| buf := make([]byte, 0, len(SqsPartitionedMsgVisPrefix)+sqsKeyCapSmall) | ||
| buf = append(buf, SqsPartitionedMsgVisPrefix...) | ||
| buf = append(buf, encodeSQSSegment(queueName)...) | ||
|
Contributor
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. |
||
| buf = append(buf, sqsPartitionedQueueTerminator) | ||
| buf = appendU32(buf, partition) | ||
| buf = appendU64(buf, gen) | ||
| return buf | ||
| } | ||
|
|
||
| // sqsPartitionedMsgDedupKey builds the FIFO dedup key for a | ||
| // partitioned queue. The dedup window is per-partition by design | ||
| // (DeduplicationScope=messageGroup with PartitionCount>1) — the | ||
| // validator in adapter/sqs_partitioning.go rejects the queue-scoped | ||
| // scope on partitioned queues, so this key shape is always reachable | ||
| // from the same partition that ran the dedup check. | ||
| func sqsPartitionedMsgDedupKey(queueName string, partition uint32, gen uint64, dedupID string) []byte { | ||
| buf := make([]byte, 0, len(SqsPartitionedMsgDedupPrefix)+sqsKeyCapLarge) | ||
| buf = append(buf, SqsPartitionedMsgDedupPrefix...) | ||
| buf = append(buf, encodeSQSSegment(queueName)...) | ||
| buf = append(buf, sqsPartitionedQueueTerminator) | ||
| buf = appendU32(buf, partition) | ||
| buf = appendU64(buf, gen) | ||
| buf = append(buf, encodeSQSSegment(dedupID)...) | ||
| return buf | ||
| } | ||
|
|
||
| // sqsPartitionedMsgGroupKey builds the FIFO group-lock key for a | ||
| // partitioned queue. partitionFor maps a MessageGroupId to one | ||
| // partition, so the group lock for any given group lives on exactly | ||
| // one partition — there is no cross-partition group-lock invariant | ||
| // to maintain. | ||
| func sqsPartitionedMsgGroupKey(queueName string, partition uint32, gen uint64, groupID string) []byte { | ||
| buf := make([]byte, 0, len(SqsPartitionedMsgGroupPrefix)+sqsKeyCapLarge) | ||
| buf = append(buf, SqsPartitionedMsgGroupPrefix...) | ||
| buf = append(buf, encodeSQSSegment(queueName)...) | ||
| buf = append(buf, sqsPartitionedQueueTerminator) | ||
| buf = appendU32(buf, partition) | ||
| buf = appendU64(buf, gen) | ||
| buf = append(buf, encodeSQSSegment(groupID)...) | ||
| return buf | ||
| } | ||
|
|
||
| // sqsPartitionedMsgByAgeKey builds the send-age index key for a | ||
| // partitioned queue. The reaper enumerates both the legacy and | ||
| // partitioned byage prefixes when reaping a queue (see | ||
| // sqsMsgByAgePrefixesForQueue) so a queue that is partitioned today | ||
| // — or, hypothetically, that gains partitions across a future | ||
| // migration — does not strand its old data. | ||
| func sqsPartitionedMsgByAgeKey(queueName string, partition uint32, gen uint64, sendTimestampMs int64, messageID string) []byte { | ||
| buf := make([]byte, 0, len(SqsPartitionedMsgByAgePrefix)+sqsKeyCapLarge) | ||
| buf = append(buf, SqsPartitionedMsgByAgePrefix...) | ||
| buf = append(buf, encodeSQSSegment(queueName)...) | ||
| buf = append(buf, sqsPartitionedQueueTerminator) | ||
| buf = appendU32(buf, partition) | ||
| buf = appendU64(buf, gen) | ||
| buf = appendU64(buf, uint64MaxZero(sendTimestampMs)) | ||
| buf = append(buf, encodeSQSSegment(messageID)...) | ||
| return buf | ||
| } | ||
|
|
||
| // sqsPartitionedMsgByAgePrefixForQueueAllPartitions returns the | ||
| // prefix shared by every partitioned byage entry for one queue | ||
| // (across all partitions and all generations). The reaper uses it | ||
| // alongside the legacy prefix to enumerate orphan records on a | ||
| // partitioned queue. | ||
| func sqsPartitionedMsgByAgePrefixForQueueAllPartitions(queueName string) []byte { | ||
| buf := make([]byte, 0, len(SqsPartitionedMsgByAgePrefix)+sqsKeyCapSmall) | ||
| buf = append(buf, SqsPartitionedMsgByAgePrefix...) | ||
| buf = append(buf, encodeSQSSegment(queueName)...) | ||
|
Contributor
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. |
||
| buf = append(buf, sqsPartitionedQueueTerminator) | ||
| return buf | ||
| } | ||
|
|
||
| // sqsMsgByAgePrefixesForQueue returns the {legacy, partitioned} | ||
| // prefix pair for a queue's byage records. The reaper iterates both: | ||
| // a queue created before HT-FIFO landed has only legacy entries; a | ||
| // partitioned queue created after PR 5 has only partitioned entries; | ||
| // no queue has both today, but enumerating both keeps the reaper | ||
| // future-proof against an offline-rebuild migration that produces a | ||
| // mixed-prefix queue. | ||
| func sqsMsgByAgePrefixesForQueue(queueName string) [][]byte { | ||
| return [][]byte{ | ||
| sqsMsgByAgePrefixAllGenerations(queueName), | ||
| sqsPartitionedMsgByAgePrefixForQueueAllPartitions(queueName), | ||
| } | ||
| } | ||
|
|
||
| // sqsPartitionedByAgeKeyHeaderLen is the byte length of the | ||
| // (partition, gen, ts) header that follows the queue segment in a | ||
| // partitioned byage key — one big-endian uint32 plus two big-endian | ||
| // uint64s. | ||
| const sqsPartitionedByAgeKeyHeaderLen = 4 + 8 + 8 | ||
|
|
||
| // parseSqsPartitionedMsgByAgeKey reverses sqsPartitionedMsgByAgeKey. | ||
| // Returns ok=false when the key does not match the expected partitioned | ||
| // shape. The reaper tries this parser when parseSqsMsgByAgeKey fails, | ||
| // so it can handle a queue with both legacy and partitioned entries | ||
| // (today only one or the other applies, but the dual-parse keeps the | ||
| // reaper safe against future migrations). | ||
| func parseSqsPartitionedMsgByAgeKey(key []byte, queueName string) (sqsPartitionedMsgByAgeRecord, bool) { | ||
| expected := sqsPartitionedMsgByAgePrefixForQueueAllPartitions(queueName) | ||
| if !bytes.HasPrefix(key, expected) { | ||
| return sqsPartitionedMsgByAgeRecord{}, false | ||
| } | ||
| rest := key[len(expected):] | ||
| if len(rest) < sqsPartitionedByAgeKeyHeaderLen { | ||
| return sqsPartitionedMsgByAgeRecord{}, false | ||
| } | ||
| partition := binary.BigEndian.Uint32(rest[:4]) | ||
| gen := binary.BigEndian.Uint64(rest[4:12]) | ||
| tsRaw := binary.BigEndian.Uint64(rest[12:sqsPartitionedByAgeKeyHeaderLen]) | ||
| msgIDEnc := string(rest[sqsPartitionedByAgeKeyHeaderLen:]) | ||
| msgID, err := decodeSQSSegment(msgIDEnc) | ||
| if err != nil { | ||
| return sqsPartitionedMsgByAgeRecord{}, false | ||
| } | ||
| // Wall-clock millis fits in int63 — see parseSqsMsgByAgeKey for | ||
| // the same bound. | ||
| if tsRaw > 1<<63-1 { | ||
| return sqsPartitionedMsgByAgeRecord{}, false | ||
| } | ||
| return sqsPartitionedMsgByAgeRecord{ | ||
| Partition: partition, | ||
| Generation: gen, | ||
| SendTimestampMs: int64(tsRaw), | ||
| MessageID: msgID, | ||
| }, true | ||
| } | ||
|
|
||
| // sqsPartitionedMsgByAgeRecord is the parsed shape of a partitioned | ||
| // byage key. Mirrors sqsMsgByAgeRecord with partition added. | ||
| type sqsPartitionedMsgByAgeRecord struct { | ||
| Partition uint32 | ||
| Generation uint64 | ||
| SendTimestampMs int64 | ||
| MessageID string | ||
| } | ||
|
|
||
| // appendU32 mirrors appendU64 for the partition segment. | ||
| func appendU32(dst []byte, v uint32) []byte { | ||
| var buf [4]byte | ||
| binary.BigEndian.PutUint32(buf[:], v) | ||
| return append(dst, buf[:]...) | ||
| } | ||
Oops, something went wrong.
Oops, something went wrong.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
The partitioned key layout is missing a delimiter after the variable-length queue name segment. Without a delimiter (like '|'), a prefix scan for a queue name that is a prefix of another queue name (e.g., 'queue' and 'queue1') will incorrectly return records from both queues. This can lead to state inconsistencies. The design document (§3.1) explicitly mentions that segments should be terminated by a '|' to prevent such collisions.
This issue applies to all partitioned key constructors in this file (sqsPartitionedMsgDataKey, sqsPartitionedMsgVisKey, sqsPartitionedMsgDedupKey, sqsPartitionedMsgGroupKey, sqsPartitionedMsgByAgeKey) and their corresponding prefix helpers.
References