Struct SyncPushBody 

pub struct SyncPushBody {
    pub sender_agent_id: String,
    pub sender_clock: VectorClock,
    pub memories: Vec<Memory>,
    pub deletions: Vec<String>,
    pub archives: Vec<String>,
    pub restores: Vec<String>,
    pub links: Vec<MemoryLink>,
    pub pendings: Vec<PendingAction>,
    pub pending_decisions: Vec<PendingDecision>,
    pub namespace_meta: Vec<NamespaceMetaEntry>,
    pub namespace_meta_clears: Vec<String>,
    pub dry_run: bool,
}

Request body for POST /api/v1/sync/push.

Fields

sender_agent_id: String

Claimed agent_id of the peer pushing data. Recorded in sync_state for vector clock advancement. Treated as identity only (not attestation) — same NHI model as every other write.

sender_clock: VectorClock

Vector clock the sender had at push time. Foundation accepts it and stores the latest-seen timestamp; full clock reconciliation lands with Task 3a.1.

memories: Vec<Memory>

Memories the sender is offering. Applied via the existing timestamp-aware merge (insert_if_newer).

deletions: Vec<String>

Memory IDs the sender has deleted and wants propagated. Applied via db::delete. v0.6.0.1: simple remove (no tombstone row); a concurrent newer insert_if_newer from another peer could revive the row — a Last-Writer-Wins quirk we live with until v0.7’s CRDT-lite tombstone table lands. In the common 4-node mesh, the same delete reaches every peer well before any revival window.

archives: Vec<String>

v0.6.2 (S29): memory IDs the sender has explicitly archived and wants propagated. Applied via db::archive_memory — a soft move from memories to archived_memories. Missing-on-peer IDs no-op. Distinct from deletions, which is a hard DELETE.

restores: Vec<String>

v0.6.2 (S29): memory IDs the sender has restored from archive and wants propagated. Applied via db::restore_archived — moves the row from archived_memories back into memories. The inverse of archives. Missing-on-peer IDs (no row in the peer’s archive table, or a live row already exists) no-op so replays are safe.

links: Vec<MemoryLink>

v0.6.2 (#325): memory links the sender wants propagated. Applied via db::create_link on each peer. Duplicates are a no-op thanks to the unique (source_id, target_id, relation) constraint on memory_links.

pendings: Vec<PendingAction>

v0.6.2 (S34): pending-action rows the sender wants propagated. Applied via db::upsert_pending_action — preserves the originator’s id + status + approvals so the cluster agrees on pending state. Without this, POST /api/v1/pending/{id}/approve on a peer 404s because the row only exists on the originator.

pending_decisions: Vec<PendingDecision>

v0.6.2 (S34): pending-action decisions the sender wants propagated so approve/reject on any node lands consistently. Applied via db::decide_pending_action — already-decided rows no-op, replay-safe.

namespace_meta: Vec<NamespaceMetaEntry>

v0.6.2 (S35): namespace-standard meta rows the sender wants propagated. Applied via db::set_namespace_standard(conn, ns, standard_id, parent.as_deref()) so the peer’s inheritance-chain walk uses the originator’s explicit parent (not a locally auto-detected one).

namespace_meta_clears: Vec<String>

v0.6.2 (S35 follow-up): namespaces whose standard the sender has cleared and wants propagated. Applied via db::clear_namespace_standard — missing-on-peer namespaces no-op so replays are safe. Without this, alice clearing a standard on node-1 left the row visible on node-2’s peer, breaking cross-peer rule-lifecycle assertions.

dry_run: bool

Preview mode — classify and count, do not write.

Trait Implementations

impl<'de> Deserialize<'de> for SyncPushBody

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.