Repository navigation
0.22.0
minanode 0.22.0 release notes
0.22.0 speeds up web node sync from distant relays, the web node's start and its
block prover. It adds robustness fixes for staged ledger data from peers and
from the frontier store. It is a beta release. The web node runs on devnet only.
Use only devnet keys in it.
Since 2026-11-04, minaweb.org and its two relays run the code in this release.
Sync from distant relays
A browser far from its relays synced slowly. Five changes address it. The
numbers below come from lab runs: a test relay, or two on one VM, with a
headless Chrome browser at a real or emulated distance. They are not guarantees.
- Larger SCTP window. A relay's WebRTC connections now let 1 MiB be in
flight instead of 256 KiB, set with the new--webrtc-sctp-window. From a
browser 165 ms away, the median sync fell from 163 s to 139 s. At 312 ms it
fell from 276 s to 209 s (#22). - More requests in flight. A relay takes 16 sync requests at once from a
browser instead of 5, and says so in its WebRTC answer. With the 1 MiB window,
the median sync from 172 ms fell from 142 s to 99 s. From 313 ms it fell from
212 s to 126 s (#24). - Smaller staged ledger. The relay sends the scan state without the proofs
of completed merge jobs, as OCaml nodes do. The staged ledger hash is
unchanged. A devnet staged ledger fell from 81 MB to 36 MB, and the median
sync at 165 ms from 215 s to 169 s (#25). - Two relays at once. The browser fetches the staged ledger from two relays
at once, one from each end, and joins the parts where they meet. With both
relays 165 ms away, the median sync fell from 131 s to 102 s. At 350 ms it
fell from 265 s to 195 s. These runs measured an earlier build of the change,
before its final cancellation and retry fixes, and were not repeated on the
final code (#39). - Retransmission after a timeout. The relay's SCTP retransmits every
outstanding chunk after a retransmission timeout, as RFC 9260 describes. The
lab runs showed no measurable change in sync time (#23).
On minaweb.org, from the US West Coast, a fresh headless Chrome reached Synced
in 70 s and 74 s with this release's page and relays, on 2026-11-04. Two earlier
checks from the same machine, with the previous page, took 178 s and 157 s, on
2026-11-03 and 2026-11-04. Timing ran from page open to Synced.
Pages and relays of different versions work together. A 0.21.0 relay gets 5
requests at a time and a one-relay staged ledger fetch from a 0.22.0 page. A
0.21.0 page sends a 0.22.0 relay at most 5 requests.
Faster start
The web node now makes its verifiers' SRS on its thread pool while its files
download (#26). It asks for those files together (#27). It sends its WebRTC
offer without the extra round trip of a CORS preflight (#28). In lab runs on
an 8-core VM, the first two changes cut the time from Start to the first peer on
screen from 4.0 s to 1.5 s near the relays. With 130 ms added each way, it fell
from 5.3 s to 3.0 s, and the third change saved about 0.2 s more.
Each dial now costs one signalling request instead of two. The minaweb.org
relays limit those requests per address, so a visitor who reloads several times
in a minute meets the limit later.
Before Start, the start screen now checks that the browser can run the node. If
WebRTC or WebAssembly is off, or the browser lacks cross-origin isolation or
shared memory, Start is off and the screen names the cause (#30).
Faster block proofs in the browser
These numbers come from lab runs of the isolated prover in headless Chrome on an
8-core VM, proving the Mesa genesis block.
- Field multiplication. WebAssembly has no 64-bit multiply with a 128-bit
result. The browser build now multiplies field elements on nine 29-bit limbs.
The median block proof fell from 15.9 s to 10.2 s (#14). - Threads. The prover uses one thread per core, at most 8, instead of
leaving one core to the node. On 8 cores the median proof fell from 11.1 s to
9.1 s (#15). - Setup. The prover reads its Lagrange bases from a cache and checks them
against the SRS, instead of computing them. Prover setup took 23 to 28 s
before. With the cache it took 8 to 9.5 s, and less than 7 s with the faster
multiplication as well (#16). - Witness. Four changes cut witness generation from 1.20 s to 0.57 s per
proof, and from 1.03 s to 0.44 s with the faster multiplication. These are
medians of later proofs, with the prover's SRS kept between proofs as the web
node keeps it. The witnesses and proofs are unchanged, byte for byte (#17,
#18, #19, #20).
With the faster prover, the web node starts producing a block 10 s before its
slot instead of 20 s, so the block builds on a newer best tip (#21). A slower
machine may finish a few seconds into its slot. Native nodes keep 6 s.
A native node that proves also uses the Lagrange cache. It writes the bases it
computes next to its circuit files, in ~/.mina/circuit-blobs/<circuits>/, and
reads them back after the same SRS check. A cache file that fails the check is
never used. mina internal check-lagrange-cache checks a cache file.
Robustness
- The node checks a pending-coinbase tree from a peer's staged ledger parts, or
from its frontier store, against the tree's leaves, as OCaml builds it. It
refuses a tree that differs (#11). - A web node whose block, transaction or zkApp proof fails verification no
longer panics while it writes a debug dump. The verification fails, as before
(#12). - If a peer's staged ledger parts match the hash but fail to rebuild, the node
rejects that peer's answer and asks another peer, as OCaml does. It panicked
before. Parts from peers and from the frontier store must also have the
scan-state shape the node builds (#13).
Web node
- The start screen lists the six start-up steps, each with a figure from the
build. After Start, the same list shows progress (#3). - The header's best tip keeps up with the page (#4). Its status chip says
Stale once/statushas not answered for 20 s (#6). - The Send page keeps the payments the tab sent across a role change and a
reload. It keeps no key or memo (#9). - The Overview and Start-up pages say what a reload does with the saved chain:
Not saved yet, Saved at #N, or a reason it syncs again (#10). - A page can read an account from the best tip's ledger with
ledger().latest().account(publicKey), together with the block it came from
(#31).
Run a relay
--webrtc-sctp-window sets the SCTP window of every rustrtc WebRTC connection a
native node makes. It is also MINA_WEBRTC_SCTP_WINDOW, and
WEBRTC_SCTP_WINDOW in the relay kit's relay.env. The default is 1 MiB.
262144 restores the 0.21.0 window. The setting counts bytes in flight and
queued. It is not a memory limit. Native libp2p connections are unchanged
(#22).
A 0.22.0 relay serves the staged ledger from either end and takes 16 sync
requests at a time from a 0.22.0 page. Upgrade relays before pages to give
browsers the full benefit. Either order works.
Web archive
mina-web-0.22.0.tar.gz runs the same web node from your own computer. Check it
against the signed SHA256SUMS, unpack it and run python3 -B serve.py. Then
open http://127.0.0.1:8080/ in desktop Chrome or Chromium. Its packaged relays
are relay1.minaweb.org and relay2.minaweb.org.
Run the web node yourself has the
steps.
Tests and tooling
mina-node-testingbuilds again, and 10 scenario suites pass (#36, #38).
The record and replay suites fail on a recorder bug, which is not fixed yet.- The heartbeat signing test expects the transaction pool's current fields
(#35). - The uptime release gate rejects equal HTTP and libp2p ports, and its fixture
check validates the fixture pair first (#32). WASM_ENCODING=brlets a minaweb deployment upload the build's brotli WASM.
It is off by default (#29).- The CHANGELOG and the 0.20.1 and 0.21.0 notes drop PR numbers from the
repository these releases were prepared in (#33, #37). The release docs
describe oneSHA256SUMSfor the native and web archives (#34).
Known issues
- Only desktop Chrome and Chromium are tested. Other browsers are unsupported.
- Relay capacity is unmeasured. So are the 1 MiB window's fairness on a shared
uplink and a relay's memory with many browsers downloading at once. - Sync over lossy paths is slow. In lab runs with 0.5% random loss at 165 ms,
builds without the two-relay fetch took about 15 minutes to reach Synced. An
earlier build of the two-relay fetch took about 17 minutes at 350 ms and 0.5%
loss. The final code was not measured with loss. - If one relay stops during the two-relay staged ledger download, the other
finishes it alone, at its own speed. The parts the stopped relay sent are
lost. - A relay that is full can drop a browser that is still syncing, to make room.
The browser then starts its download again. - Worker purchases and staked block production from the web node are not
qualified on public devnet. The faster prover produced blocks in a browser tab
on a private Mesa chain (#14). - While a relay's host is down, its TURN server stays in the tab's list until
its credentials expire. Each new connection can then take up to 3 s longer. - The relays offer TURN over UDP and TCP on port 3478, not over TLS on port 443.
A network that allows only HTTPS can't reach them.
Checksums
SHA256SUMS lists the release archives:
109b971dc4d26e6387bbb48b5bfd98c492f91a682d15eb6c5062eea3feed4a78 mina-0.22.0-linux-x86_64.tar.gz
23c52c369897eb89091a8ae93451ddea35e335265daea33f80f4c4bd489aa4e7 mina-0.22.0-linux-arm64.tar.gz
1ea4362719931eaecbb2164ab8b5b4e596c7dac7eff44bea7b1b40e1c575a54d mina-web-0.22.0.tar.gz
Docker: ghcr.io/minanode/mina-rust:0.22.0
(ghcr.io/minanode/mina-rust@sha256:1df8bf7baddb0dca947fb46ddd603b83cdf3a1588dc8af7f9eda1103c27d1729)
SHA256SUMS.sig is the SSH signature of SHA256SUMS by release@minanode.org.
Obtain allowed_signers from minanode.org and compare its fingerprint with the
install guide. Verify before
unpacking:
ssh-keygen -Y verify -f allowed_signers -I release@minanode.org \
-n file -s SHA256SUMS.sig < SHA256SUMS &&
sha256sum --check --ignore-missing SHA256SUMS