1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
//! `RENAME` / `RENAMENX` orchestration. The runtime decides whether
//! both keys live on the same shard (atomic single-Op route) or split
//! across shards (Take-Put orchestrator landing in v2-3b — until then
//! the cross-shard arm returns `-CROSSSHARD ...`).
//!
//! Why this can't be served by `Route::Single`: a single-key route hits
//! the source shard, where a same-shard atomic is straightforward but a
//! cross-shard hop needs to know nshards + emit a different Op to ship
//! the value to the destination shard. Routing is the runtime's job;
//! the dispatch layer (`kevy::cmd::*`) sees only one shard at a time.
use crate::message::{Agg, Inbound, Op, Part, PendingSlot, RenameStep};
use crate::reduce::{drain_front, shard_of};
use crate::shard::Shard;
use crate::{Commands, RespVersion};
use kevy_resp::ArgvView;
impl<C: Commands> Shard<C> {
/// `RENAME` / `RENAMENX` — see [`Route::Rename`].
pub(crate) fn start_rename<A: ArgvView + ?Sized>(
&mut self,
conn_id: u64,
seq: u64,
args: &A,
nx: bool,
) {
// Arity: RENAME source destination → 3 args.
if args.len() != 3 {
let cmd_name = if nx { "renamenx" } else { "rename" };
let err = format!("-ERR wrong number of arguments for '{cmd_name}' command\r\n");
self.fold_rename_reply(conn_id, seq, err.into_bytes());
return;
}
let src = args[1].to_vec();
let dst = args[2].to_vec();
let src_shard = shard_of(&src, self.nshards);
let dst_shard = shard_of(&dst, self.nshards);
if src_shard == dst_shard {
// Same-shard: one atomic Op::Rename. Route to the owning
// shard — exec_op runs store.rename + bumps WATCH versions
// + AOF logs + emits keyspace notifications.
self.push_pending_slot(conn_id, 1, Agg::First(None), false);
let op = Op::Rename { src, dst, nx };
if src_shard == self.id {
let part = self.exec_op(op);
self.fold(conn_id, seq, part);
} else {
self.send_to(
src_shard,
Inbound::Request {
origin: self.id,
conn: conn_id,
seq,
op,
},
);
}
return;
}
// Cross-shard: orchestrator. Push a single pending slot with
// Agg::RenameOrchestrator; step 1 emits Op::RenameTake to
// src_shard. Fold receives Part::RenameTaken (or NoSuchSrc),
// step transitions to Put, emits Op::RenamePut. Step 2's
// Part::RenamePutDone triggers the +OK / :1 / :0 reply.
let agg = Agg::RenameOrchestrator {
step: RenameStep::Take,
nx,
src: src.clone(),
dst,
dst_shard,
taken: None,
put_stored: None,
};
if let Some(c) = self.conns.get_mut(&conn_id) {
let proto = c.proto;
c.pending.push_back(PendingSlot {
remaining: 1,
agg,
done: None,
proto,
});
}
let take_op = Op::RenameTake(src);
if src_shard == self.id {
let part = self.exec_op(take_op);
self.fold(conn_id, seq, part);
} else {
self.send_to(
src_shard,
Inbound::Request {
origin: self.id,
conn: conn_id,
seq,
op: take_op,
},
);
}
}
/// Resume the cross-shard RENAME after a sub-reply lands. Called
/// from `Shard::fold` once `slot.remaining == 0` for an
/// `Agg::RenameOrchestrator` slot.
///
/// On step-1 completion: if Take succeeded → ship step 2 to
/// dst_shard, re-arm the slot. If Take missed → finalize with
/// `-ERR no such key`.
///
/// On step-2 completion: finalize with `+OK` (RENAME ok) or `:1`
/// (RENAMENX ok) or `:0` (RENAMENX-blocked: dst already existed
/// on dst_shard at the moment of Put; we accept the data-loss
/// race vs adding a third "restore-src" step — Redis cluster has
/// the same trade-off via MIGRATE).
pub(crate) fn finalize_rename_agg(&mut self, conn_id: u64, seq: u64, agg: Agg) {
let Agg::RenameOrchestrator {
step,
nx,
src,
dst,
dst_shard,
taken,
put_stored,
} = agg
else {
return;
};
match step {
RenameStep::Take => self.advance_rename_to_put(conn_id, seq, nx, src, dst, dst_shard, taken),
RenameStep::Put => self.finish_rename_put(conn_id, seq, nx, put_stored),
}
}
/// Step 1 → step 2 transition. If src didn't exist, finalize with
/// `-ERR no such key`. Otherwise re-arm the slot for Put + ship
/// Op::RenamePut to dst_shard.
#[allow(clippy::too_many_arguments)]
fn advance_rename_to_put(
&mut self,
conn_id: u64,
seq: u64,
nx: bool,
_src: Vec<u8>,
dst: Vec<u8>,
dst_shard: usize,
taken: Option<(kevy_store::Value, Option<u64>)>,
) {
let Some((value, ttl_ms)) = taken else {
// Take returned NoSuchSrc — finalize.
self.fill_rename_slot(conn_id, seq, b"-ERR no such key\r\n".to_vec());
return;
};
// Re-arm the same slot for step 2. Step transitions to Put.
if let Some(c) = self.conns.get_mut(&conn_id) {
let idx = (seq - c.next_emit) as usize;
if let Some(slot) = c.pending.get_mut(idx) {
slot.remaining = 1;
slot.agg = Agg::RenameOrchestrator {
step: RenameStep::Put,
nx,
src: Vec::new(), // src no longer needed
dst: dst.clone(),
dst_shard,
taken: None, // step 2 doesn't buffer Value
put_stored: None, // fold fills this from RenamePutDone
};
}
}
let put_op = Op::RenamePut {
dst,
value,
ttl_ms,
nx,
};
if dst_shard == self.id {
let part = self.exec_op(put_op);
self.fold(conn_id, seq, part);
} else {
self.send_to(
dst_shard,
Inbound::Request {
origin: self.id,
conn: conn_id,
seq,
op: put_op,
},
);
}
}
/// Step 2 finished. Reply +OK / :1 / :0 depending on Put's `stored`
/// flag + the NX flag. `put_stored` is filled by `Shard::fold` from
/// `Part::RenamePutDone.stored` before this is called.
fn finish_rename_put(
&mut self,
conn_id: u64,
seq: u64,
nx: bool,
put_stored: Option<bool>,
) {
let stored = put_stored.unwrap_or(false);
let reply = if stored {
if nx { b":1\r\n".to_vec() } else { b"+OK\r\n".to_vec() }
} else {
// Only reachable on RENAMENX cross-shard with pre-existing
// dst at Put time. The data-loss race here (src already
// taken at this point — Put refused, src gone) is
// documented. Future v3 could add a restore-src step or
// 2-phase pre-check.
b":0\r\n".to_vec()
};
self.fill_rename_slot(conn_id, seq, reply);
}
/// Drop a literal RESP frame into the orchestrator's slot + drain.
fn fill_rename_slot(&mut self, conn_id: u64, seq: u64, bytes: Vec<u8>) {
if let Some(c) = self.conns.get_mut(&conn_id) {
let idx = (seq - c.next_emit) as usize;
if let Some(slot) = c.pending.get_mut(idx) {
slot.done = Some(bytes);
}
drain_front(c);
}
}
/// Push a single pending slot + immediately fold a pre-built reply
/// into it. Used by the synchronous error paths (arity check,
/// cross-shard rejection) so the reply preserves seq order without
/// going through the broader `immediate_reply` (which assigns a
/// fresh seq — we already have one).
fn fold_rename_reply(&mut self, conn_id: u64, seq: u64, reply: Vec<u8>) {
if let Some(c) = self.conns.get_mut(&conn_id) {
let proto = c.proto;
c.pending.push_back(PendingSlot {
remaining: 1,
agg: Agg::First(None),
done: None,
proto,
});
// Silence unused — proto only matters for the few aggs that
// care about RESP3 shape, not for a fixed-bytes reply.
let _ = RespVersion::V2;
}
self.fold(conn_id, seq, Part::Reply(reply));
}
}