add tsdb storage engine (& many misc changes in support)

Author: agavra

AGENTS.md

@@ -65,6 +65,25 @@ fn should_return_error_when_key_not_found() {
 - Use `BytesMut` for mutable byte buffers during encoding
 - The `bytes` crate is available as a workspace dependency
 
+### Imports
+
+- Place all `use` statements at the module level, not inside functions or methods
+- Group imports logically (standard library, external crates, local modules)
+
+```rust
+// Good
+use bytes::{BufMut, BytesMut};
+
+fn some_function() {
+    let mut buf = BytesMut::new();
+}
+
+// Bad
+fn some_function() {
+    use bytes::BufMut;  // Don't do this
+}
+```
+
 ### Dependencies
 
 - **slatedb**: The underlying storage engine for all database implementations

Cargo.lock

@@ -11,11 +11,12 @@ version = "0.1.0"
 edition = "2024"
 
 [workspace.dependencies]
-slatedb = "0.9.1"
-opendata-common = { path = "./opendata-common" }
-rstest = "0.19"
+async-trait = "0.1"
 bytes = "1.0"
 chrono = "0.4"
+dashmap = "6.1.0"
+opendata-common = { path = "./opendata-common" }
+rstest = "0.19"
+slatedb = "0.9.1"
 tokio = { version = "1.0", features = ["full"] }
-async-trait = "0.1"
 tracing = "0.1"

Cargo.toml

@@ -4,17 +4,19 @@ version.workspace = true
 edition.workspace = true
 
 [dependencies]
-slatedb.workspace = true
-opendata-common.workspace = true
+async-trait.workspace = true
 bytes.workspace = true
-roaring = "0.7"
-tsz = "0.1"
-blake3 = "1.8.2"
 chrono.workspace = true
-promql-parser = "0.6"
+dashmap.workspace = true
+opendata-common.workspace = true
+slatedb.workspace = true
 tokio.workspace = true
-async-trait.workspace = true
 tracing.workspace = true
 
+blake3 = "1.8.2"
+promql-parser = "0.6"
+roaring = "0.7"
+tsz = "0.1"
+
 [dev-dependencies]
 rstest.workspace = true

open-tsdb/Cargo.toml

@@ -158,31 +158,26 @@ execution to map series IDs back to their label sets when needed.
 - `time_bucket` (u32): The number of minutes since the UNIX epoch
 - `series_fingerprint` (u128): The fingerprint of the label set, computed as a hash of the labels
 
-The series fingerprint is computed using the labels of the series, which means
-there is a small chance that two series with different labels will have the same
-fingerprint. Prometheus makes the same trade-off and accepts the risk of
-collisions. The value contains the series ID(s) that correspond to this
-fingerprint, typically a single series ID. Note that these IDs are scoped to a
-single time bucket so the cardinality is measured in the number of series in the
-bucket, not globally.
+The series fingerprint is computed using the labels of the series. The value
+contains the series ID that corresponds to this fingerprint. Note that these IDs
+are scoped to a single time bucket so the cardinality is measured in the number
+of series in the bucket, not globally.
 
 **Value Schema:**
 
 ```
 ┌──────────────────────────────────────────────────────────┐
 │                   SeriesDictionaryValue                  │
 ├──────────────────────────────────────────────────────────┤
-│  FixedElementArray<series_id: u32>                             │
+│  series_id: u32                                          │
 └──────────────────────────────────────────────────────────┘
 ```
 
 **Structure:**
 
-- The value is a `FixedElementArray<u32>` encoding the list of `series_id` values that
-  have the specified fingerprint. In the common case of no collisions, this array
-  contains a single `series_id`. When collisions occur, multiple series IDs are
-  stored, and the forward index must be consulted to disambiguate by comparing
-  the actual label sets.
+- The value is a single `series_id` (u32) that has the specified fingerprint.
+  Hash collisions are assumed not to occur (we should consider changing this
+  to a FixedElementArray<SeriesId> to allow for hash collisions).
 
 ### `ForwardIndex` (`RecordType::ForwardIndex` = `0x03`)