solana/rfcs/0004-leader-rotation.md

Leader Rotation

The goal of this RFC is to define how leader nodes are rotated in Solana, how rotation may cause forks to arise, and how the converges in response.

Leader Seed Generation

Leader selection is decided via a random seed. The process is as follows:

1. Periodically at a specific PoH tick count select the first vote signatures that create a supermajority from the previous voting round.
2. Append them together.
3. Hash the string for N counts via a similar process as PoH itself.
4. The resulting hash is the random seed for M counts, M leader periods, where M > N

Leader Rotation

1. The leader is chosen via a random seed generated from stake weights and votes (the leader schedule)
2. The leader is rotated every T PoH ticks (leader period), accoding to the leader schedule
3. The schedule is applicable for M voting rounds

Leader's transmit for a count of T PoH ticks. When T is reached all the validators should switch to the next scheduled leader. To schedule leaders, the supermajority + M nodes are shuffled using the above calculated random seed.

All T ticks must be observed from the current leader for that part of PoH to be accepted by the network. If T ticks (and any intervening transactions) are not observed, the network optimistically fills in the T ticks, and continues with PoH from the next leader.

Partitions, Forks

Forks can arise at PoH tick counts that correspond to leader rotations, because leader nodes may or may not have observed the previous leader's data. These empty ticks are generated by all nodes in the network at a network-specified rate for hashes/per/tick Z.

There are only two possible versions of the PoH during a voting period: PoH with T ticks and entries generated by the current leader, or PoH with just ticks. The "just ticks" version of the PoH can be thought of as a virtual ledger, one that all nodes in the network can derive from the last tick in the previous period.

Validators can ignore forks at other points (e.g. from the wrong leader), or slash the leader responsible for the fork.

Validators vote on the longest chain that contains their previous vote, or a longer chain if the lockout on their previous vote has expired.

Validator's View

Time Progression

The diagram below represents a validator's view of the PoH stream with possible forks over time. L1, L2, etc. are leader periods, and Es represent entries from that leader during that leader's period. The 'x's represent ticks only, and time flows downwards in the diagram.

time  +----+                                          validator action
|     | L1 |                 E(L1)
|     |----|            /             \                vote(E(L2))
|     | L2 |        E(L2)               x
|     |----|        /    \           /     \           vote(E(L2))
|     | L3 |    E(L3)      x       E(L3)'    x
|     |----|    /  \     /   \     /  \     /  \       slash(L3)
|     | L4 |  x     x  E(L4)  x   x    x   x    x
V     |----|  |     |  |      |   |    |   |    |      vote(E(L4))
V     | L5 |  xx   xx  xx   E(L5) xx   xx  xx   xx
V     +----+                                           hang on to E(L4) and E(L5) for more...

Note that an E appearing on 2 branches at the same period is a slashable condition, so a validator observing E(L3) and E(L3)' can slash L3 and safely choose x for that period. Once a validator observes a supermajority vote on any branch, other branches can be discarded below that tick count. For any period, validators need only consider a single "has entries" chain or a "ticks only" chain.

Time Division

It's useful to consider leader rotation over PoH tick count as time division of the job of encoding state for the network. The following table presents the above tree of forks as a time-divided ledger.

leader period	L1	L2	L3	L4	L5
data	E(L1)	E(L2)	E(L3)	E(L4)	E(L5)
ticks to prev				x	xx

Note that only data from leader L3 will be accepted during leader period L3. Data from L3 may include "catchup" ticks back to a period other than L2 if L3 did not observe L2's data. L4 and L5's transmissions include the "ticks to prev" PoH entries.

This arrangement of the network data streams permits nodes to save exactly this to the ledger for replay, restart, and checkpoints.

Leader's View

When a new leader begins a period, it must first transmit any PoH (ticks) required to link the new period with the most recently observed and voted period.

Examples

Small Partition

1. Network partition M occurs for 10% of the nodes
2. The larger partition K, with 90% of the stake weight continues to operate as normal
3. M cycles through the ranks until one of them is leader, generating ticks for periods where the leader is in K.
4. M validators observe 10% of the vote pool, finality is not reached.
5. M and K re-connect.
6. M validators cancel their votes on M, which has not reached finality, and re-cast on K (after their vote lockout on M).

Leader Timeout

1. Next rank leader node V observes a timeout from current leader A, fills in A's period with virtual ticks and starts sending out entries.
2. Nodes observing both streams keep track of the forks, waiting for: a. their vote on leader A to expire in order to be able to vote on B b. a supermajority on A's period
3. If a occurs, leader B's period is filled with ticks, if b occurs, A's period is filled with ticks
4. Partition is resolved just like in the Small Partition

Network Variables

M - number of nodes outside the supermajority to whom leaders broadcast their PoH for validation

N - number of voting rounds for which a leader schedule is considered before a new leader schedule is used

T - number of PoH ticks per leader period (also voting period)

Z - number of hashes per PoH tick