Write HNSW checkpoints for all vector indexes to disk. Called periodically from the TPC event loop (e.g., every 5 minutes or when idle). Each index is serialized to a file at `{data_dir}/vector-ckpt/{index_key}.ckpt`. After checkpointing, WAL replay only needs to process entries since the checkpoint — not the entire history.
(&self)
| 56 | /// After checkpointing, WAL replay only needs to process entries |
| 57 | /// since the checkpoint — not the entire history. |
| 58 | pub fn checkpoint_vector_indexes(&self) -> usize { |
| 59 | if self.vector_collections.is_empty() { |
| 60 | return 0; |
| 61 | } |
| 62 | |
| 63 | let ckpt_dir = self.data_dir.join("vector-ckpt"); |
| 64 | if std::fs::create_dir_all(&ckpt_dir).is_err() { |
| 65 | tracing::warn!( |
| 66 | core = self.core_id, |
| 67 | "failed to create vector checkpoint dir" |
| 68 | ); |
| 69 | return 0; |
| 70 | } |
| 71 | |
| 72 | let mut checkpointed = 0; |
| 73 | for (key, collection) in &self.vector_collections { |
| 74 | if collection.is_empty() { |
| 75 | continue; |
| 76 | } |
| 77 | let bytes = collection.checkpoint_to_bytes(self.vector_checkpoint_kek.as_ref()); |
| 78 | if bytes.is_empty() { |
| 79 | continue; |
| 80 | } |
| 81 | // Checkpoint filename uses the old-style `"{tid}:{coll}"` form so |
| 82 | // existing on-disk checkpoint files remain valid across upgrades. |
| 83 | let filename = CoreLoop::vector_checkpoint_filename(key); |
| 84 | let ckpt_path = ckpt_dir.join(format!("{filename}.ckpt")); |
| 85 | let tmp_path = ckpt_dir.join(format!("{filename}.ckpt.tmp")); |
| 86 | if nodedb_wal::segment::atomic_write_fsync(&tmp_path, &ckpt_path, &bytes).is_ok() { |
| 87 | checkpointed += 1; |
| 88 | } |
| 89 | } |
| 90 | |
| 91 | if checkpointed > 0 { |
| 92 | tracing::info!( |
| 93 | core = self.core_id, |
| 94 | checkpointed, |
| 95 | total = self.vector_collections.len(), |
| 96 | "vector collections checkpointed" |
| 97 | ); |
| 98 | } |
| 99 | checkpointed |
| 100 | } |
| 101 | |
| 102 | /// Load HNSW checkpoints from disk on startup, before WAL replay. |
| 103 | /// |
no test coverage detected