Skip to content

skills: v6.6.4 and v6.7.1–v6.7.4 catch-up - #20

Draft
alexander-sei wants to merge 6 commits into
mainfrom
alexanderh/plt-1580-bring-sei-skill-up-to-date-with-current
Draft

alexander-sei wants to merge 6 commits into
mainfrom
alexanderh/plt-1580-bring-sei-skill-up-to-date-with-current

Conversation

@alexander-sei

Copy link
Copy Markdown
Contributor

Brings the skill up to date with the sei-chain patch releases that never got a catch-up PR: v6.6.4 and v6.7.1 through v6.7.4. Part of PLT-1580.

The docs bridge did not open these. It has refused v6.6.4 on every run since that release appeared: the v6.6.4 release notes link sei-protocol/sei-chain#3978 in a prose sentence, and the bridge's release-notes cross-check treats every linked PR as one the tag range must contain. So this was done by hand, one commit per change, each checked against the release tag.

Skill changes

Release sei-chain change Skill files
v6.6.4, v6.7.2 sei-protocol/sei-chain#4448, sei-protocol/sei-chain#4447: a frozen node's EVM RPC comes back after a restart. Added which release a frozen node runs (v6.6.4 for the v6.7 upgrade) and the v6.7 freeze heights. node-operations.md, seid-cli.md
v6.7.1 sei-protocol/sei-chain#4478: seid export removed seid-cli.md (new seid rollback section)
v6.7.1 sei-protocol/sei-chain#4496: seid rollback refuses targets below FlatKV's retained history seid-cli.md, common-errors.md
v6.7.1 sei-protocol/sei-chain#4484: a pinned write mode must match the persisted migration state or the node refuses to start node-operations.md, staking-governance.md
v6.7.2 sei-protocol/sei-chain#4500, sei-protocol/sei-chain#4514: /net_info n_peers counts connected peers; new tendermint_p2p_known_addresses gauge common-errors.md, validators.md
v6.7.3 sei-protocol/sei-chain#4679: the state sync trust period is built into seid (112h on pacific-1, atlantic-2, arctic-1) and trust-period in config.toml is ignored node-operations.md
— The skill said Go 1.24.x; go.mod has said 1.25.6 since v6.4.0 seid-cli.md, node-operations.md

Reviewed, no skill change

Verification

  • Error strings, flag defaults, metric names and the trust-period table read from the sei-chain source at the tag that shipped each change.
  • applied_plan/v6.7 returns 236450000 on pacific-1 and 273966500 on atlantic-2.
  • Consistent with docs.sei.io: node/frozen-nodes, node/giga-storage-migration, node/index (Go 1.25.6).
  • No reference files added or renamed, so the SKILL*.md links are unchanged.

Not in this PR

v6.8.0 is not tagged yet, but atlantic-2 applied the v6.8 upgrade at height 282577035 and runs v6.8.0-rc3. That catch-up is much larger (about 400 commits since v6.7.4) and will move the Go requirement to 1.27.3.

alexander-sei and others added 6 commits November 12, 2026 11:20
…ei-protocol/sei-chain#4448)

v6.6.4's only change: a frozen node never processes another block, and
on v6.6.3 the EVM HTTP and WS servers waited for a processed block, so
after a restart EVM JSON-RPC stayed down while Tendermint RPC kept
working. v6.6.4 (#4448) and v6.7.2 (#4447, backport #4507) initialize
the app from the last stored header at startup instead.

Add which release a frozen node runs (latest patch of the release the
upgrade replaced, never the release it introduces) and the v6.7 freeze
heights, matching docs.sei.io/node/frozen-nodes. Heights checked against
applied_plan/v6.7 on pacific-1 (236450000) and atlantic-2 (273966500).

Co-authored-by: Cursor <[email protected]>
…story (sei-protocol/sei-chain#4478, #4496)

Both shipped in v6.7.1 (backports #4489 and #4502):

- #4478 removes `seid export` and the module export paths. The root
  command now registers only start, tendermint, version and rollback
  from the server group; init, collect-gentxs and genesis import are
  unchanged. The skill no longer mentioned export, but agents reach for
  it from general Cosmos SDK knowledge, so say it is gone and what to
  use instead.
- #4496: when FlatKV is open, CompositeCommitStore.Rollback calls
  requireRollbackReachable before either backend moves and fails with
  "flatkv cannot roll back to version %d: %w; use state sync instead"
  (wrapping flatkv.ErrVersionUnreachable, "version unreachable").

Add a short seid rollback reference (--num-blocks/-n default 1, and the
height-mismatch cases from sei-cosmos/server/rollback.go at v6.7.4) as
the home for both, and name the new refusal in common-errors.md.

Co-authored-by: Cursor <[email protected]>
sei-protocol/sei-chain#4484)

Shipped in v6.7.1 (backport #4503). resolveCurrentWriteMode, which
LoadLatest and (since #4477) Rollback both reach, now calls
requirePinnedModeMatchesMigration for pins to evm_migrated,
all_migrated_but_bank and flatkv_only. When the mode derived from the
FlatKV migration metadata is neither the pin nor memiavl_only, it fails
with:

  write mode is pinned to %q but the persisted migration state is %q;
  set state-commit.sc-write-mode-enable-auto = true to follow the migration

Document the refusal, the recovery (auto mode reconciles the backends),
and the no-metadata exemption next to the existing pinning guidance in
node-operations.md and staking-governance.md. Wording matches
docs.sei.io/node/giga-storage-migration.

Co-authored-by: Cursor <[email protected]>
…known_addresses (sei-protocol/sei-chain#4500, #4514)

Both shipped in v6.7.2 (backports #4515 and #4520):

- #4500: Environment.NetInfo sets n_peers to len(peerConnections)
  instead of len(peers). Before, it counted Router.AllAddrs(), the live
  connections plus every PEX address (~175-200 on pacific-1 for nodes
  connected to ~45). `peers` still lists every known address.
  `peer_connections` already listed only live connections on v6.6.4 and
  v6.7.0, so its length is the version-independent connected count.
- #4514: new gauge tendermint_p2p_known_addresses (namespace tendermint,
  subsystem p2p) for the known-address count; tendermint_p2p_peers is
  unchanged and still counts live connections.

common-errors.md used n_peers as the connected-peer check, which over-
reported on older binaries; note the change and give the portable
query. Explain the two p2p gauges under validators.md's alert table.

Co-authored-by: Cursor <[email protected]>
…tocol/sei-chain#4679)

Shipped in v6.7.3 (backport #4690; also in v6.8.0-rc1 via #4691).
config.StateSyncTrustPeriod(chainID) returns 112h for pacific-1,
atlantic-2 and arctic-1 and 168h for any other chain, and the state sync
reactor uses it for both state providers. StateSyncConfig.TrustPeriod,
its default and its validation are gone, and trust-period is dropped
from the config.toml template. A leftover value is ignored;
warnIgnoredStateSyncTrustPeriod logs a warning at start when it is set
and differs, whether or not state sync is enabled.

The state sync recipe never set trust-period, and its trust height (the
last 100,000-block boundary, ~13h at the current ~0.48s mainnet block
time) is well inside 112h, so only add the trust-period note, the
expiry error (light.ErrOldHeaderExpired) and the warning text.

Co-authored-by: Cursor <[email protected]>
go.mod has said `go 1.25.6` since sei-chain#2713 (first released in
v6.4.0; v6.3.0 was 1.24.5) and still does at v6.6.4 and v6.7.4, so "Go
1.24.x required" has been wrong for every release since v6.4.0. A 1.24
toolchain with GOTOOLCHAIN=local refuses to build. docs.sei.io
(node/index, evm/installing-seid-cli) already says 1.25.6 or later and
points at go.mod as the authority; match it.

v6.8.0-rc3 moves to go 1.27.3; update again with the v6.8.0 catch-up.

Co-authored-by: Cursor <[email protected]>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant