You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Documents 3.7 changes the docs did not cover. Each claim was checked against this branch's source.
Configuration
--caplin.builder.allow-private-urls: lets validator-submitted builder URLs resolve to private or loopback addresses; does not apply to --caplin.mev-relay-url (cl/beacon/builder/client.go).
--http.compression: Zstandard or gzip per the client's Accept-Encoding, for responses of 1 KiB or more (node/rpcstack.go).
MAX_REORG_DEPTH: once a finalized block is known (after the initial sync cycle), reorg data is kept back to it instead; the value still caps newPayload ancestor download and side-fork validation (execution/execfinality/context.go).
RPC
erigon_getLogs, erigon_getLatestLogs, erigon_getLogsByHash: block time as blockTimestamp (was timestamp).
Fork-gated call fields: access lists before Berlin, blob hashes before Cancun, and the blob checks from Cancun on, with their errors (ValidateBlobPrerequisites).
eth_syncing during snapshot download and the new eth_subscribe("syncing").
blockTimestamp on filter and subscription logs.
callTracerwithLog: block-wide log indices; failed CREATE frames have no to.
Operations
GOMEMLIMIT defaults to 80% of the lower of RAM and the cgroup limit (FAQ Fix CircleCI #29, OOM entry).
Automatic MDBX compaction at startup (free pages ≥ 10 GB and > 4× live data).
EIP-7702 delegated senders: one in-flight transaction at the current nonce.
Branch-specific (release/3.7)
Upgrade notes: Glamsterdam on Sepolia at epoch 353024, 6 Oct 2026 13:53:36 UTC (sepolia.jsonamsterdamTime, GloasForkEpoch), and the blockTimestamp rename.
eth_syncing reports startingBlock as 0x0; the subscription reports the block it first saw syncing.
The Sepolia bullet names v3.7.1 as the minimum: #24386, #24441 and #24474 (Glamsterdam/Gloas) landed after v3.7.0. Both versions are declared with llms-pinned-version.
Same content on main: #24490. npm run build, the script tests and generate-llms.py --check pass. Pushed with --no-verify: the pre-push review flags the frozen v3.6 archive's Shutter page on every release/3.7 branch.
- --caplin.builder.allow-private-urls (cmd/utils/flags.go, added with
Gloas builder preferences): lets validator-submitted builder URLs
resolve to private or loopback addresses; other non-public ranges stay
blocked, and --caplin.mev-relay-url is unaffected
(cl/beacon/builder/client.go isAllowedBuilderIP).
- erigon_getLogs / erigon_getLatestLogs return the block time as
blockTimestamp (was timestamp, ErigonLog), matching eth_getLogs, and
erigon_getLogsByHash now includes it (rpc/jsonrpc/erigon_receipts.go).
- --http.compression serves Zstandard when the client sends
Accept-Encoding: zstd, gzip otherwise, for bodies of 1 KiB or more
(node/rpcstack.go).
- GOMEMLIMIT now defaults to 80% of the lower of RAM and the cgroup
limit unless set; GOMEMLIMIT=off leaves the heap unbounded
(node/debug/gomemlimit.go). FAQ #29 and the OOM entry said nothing of it.
…nges
- eth: fork-gated call fields (access lists before Berlin, blob hashes
before Cancun; execution/protocol/txn_executor.go preCheck),
eth_syncing during snapshot download and the new
eth_subscribe("syncing") (rpc/jsonrpc/eth_subscribe_syncing.go,
node/shards/sync_state.go), and blockTimestamp on filter and
subscription logs (rpc/rpchelper/logsfilter.go).
- debug: callTracer withLog indices are block-wide; failed CREATE
frames omit "to" (eth/tracers/native/call.go).
- txpool: EIP-7702 delegated senders get one in-flight transaction at
the current nonce (txnprovider/txpool/pool.go).
- database: automatic compaction at startup when free pages are at
least 10 GB and over 4x live data (db/kv/backup/datadir.go).
- MAX_REORG_DEPTH only bounds reorg data until a finalized block is
known (execution/execfinality/context.go).
- Upgrade notes: Glamsterdam activates on Sepolia at epoch 353024,
6 Oct 2026 13:53:36 UTC (sepolia.json amsterdamTime, clparams
GloasForkEpoch); 3.6 does not schedule it.
- Blob-hash rows apply from Cancun on: before Cancun every blob call
gets "BlobTx transactions require Cancun" (ValidateBlobPrerequisites
checks Cancun first). Add the invalid-version-byte row, and stop
presenting the method list as exhaustive.
- --http.compression follows the client's Accept-Encoding; a client
accepting neither encoding gets an uncompressed response.
- eth_syncing reports synced when fewer than 8 blocks behind; the
syncing subscription only seeds a last state if one was reported.
- MAX_REORG_DEPTH: finality is ignored during the initial sync cycle,
and the value still caps newPayload ancestor download and side-fork
validation (block_downloader.go, exec_module.go).
- txpool: delegated replacement needs the usual price bump; the trim
leaves at most one transaction.
- debug: the callTracer note is an admonition, not a method heading.
- 3.7: the syncing subscription reports the block it was at when it
first saw the node syncing, not where the sync began.
"The latest 3.7 release" is v3.7.0 today, which predates the
Glamsterdam/Gloas fixes the fork needs (#24386, #24441, #24474), so an
operator could follow the note and still run an incompatible version.
Both versions are historical, so they are declared with
llms-pinned-version: v3.7.0 is the current release on this branch and
v3.7.1 will be, and either literal would otherwise fail
test_nothing_in_the_sources_hardcodes_the_release.
…ll blobs and GOMEMLIMIT
- eth_syncing: the 8-block distance is absolute, so a node ahead of a
stale highest block reports syncing with highestBlock == currentBlock.
- eth_subscribe("syncing"): a slow client can miss intermediate states
(SendLatest on an 8-entry buffer drops the oldest).
- trace_call: TraceCallParam has no blobVersionedHashes on 3.7.
- GOMEMLIMIT example: 26GiB exceeds the 80% default on a 32 GB host; use
20GiB.
…rigontech#24490)
Documents 3.7 changes the docs did not cover. Each claim was checked
against this branch's source.
**Configuration**
- `--caplin.builder.allow-private-urls`: lets validator-submitted
builder URLs resolve to private or loopback addresses; does not apply to
`--caplin.mev-relay-url` (`cl/beacon/builder/client.go`).
- `--http.compression`: Zstandard or gzip per the client's
`Accept-Encoding`, for responses of 1 KiB or more (`node/rpcstack.go`).
- `MAX_REORG_DEPTH`: once a finalized block is known (after the initial
sync cycle), reorg data is kept back to it instead; the value still caps
newPayload ancestor download and side-fork validation
(`execution/execfinality/context.go`).
**RPC**
- `erigon_getLogs`, `erigon_getLatestLogs`, `erigon_getLogsByHash`:
block time as `blockTimestamp` (was `timestamp`).
- Fork-gated call fields: access lists before Berlin, blob hashes before
Cancun, and the blob checks from Cancun on, with their errors
(`ValidateBlobPrerequisites`).
- `eth_syncing` during snapshot download and the new
`eth_subscribe("syncing")`.
- `blockTimestamp` on filter and subscription logs.
- `callTracer` `withLog`: block-wide log indices; failed CREATE frames
have no `to`.
**Operations**
- `GOMEMLIMIT` defaults to 80% of the lower of RAM and the cgroup limit
(FAQ erigontech#29, OOM entry).
- Automatic MDBX compaction at startup (free pages ≥ 10 GB and > 4× live
data).
- EIP-7702 delegated senders: one in-flight transaction at the current
nonce.
**Branch-specific (`main`)**
- `eth_callMany` rejects `pending`; a call object with another chain's
`chainId` is rejected (erigontech#24414).
- `eth_syncing` and the subscription report the block where the sync
session began.
- The upgrade-notes bullets (`blockTimestamp`, Sepolia Glamsterdam)
follow once erigontech#24436 brings the "Upgrading from 3.6 to 3.7" section to
`main`.
Same content on `release/3.7`: erigontech#24491. `npm run build`, the script tests
and `generate-llms.py --check` pass.
---------
Co-authored-by: Bloxster <gianni.morselli@erigon.tech>
Clarify compaction requires space for a second live-data copy
docs/site/docs/fundamentals/database.md:142
CompactInPlace copies only the database's live data (db/kv/backup/backup.go:120-123), not another full copy of the bloated database. Because auto-compaction targets databases with more than 4× as much free space as live data, saying it needs the “same” space can substantially overstate the required filesystem headroom. Please describe the requirement as a second copy of the live data and regenerate the LLM outputs.
Correct compaction headroom description to reference live data
docs/site/static/llms-full.txt:1858
The compaction copy contains only live data (db/kv/backup/backup.go:120-123), not a second full copy of the bloated database. This generated paragraph therefore overstates the required filesystem headroom; update the source to say “a second copy of its live data” and regenerate this file.
Correct compaction headroom description to reference live data
llms-full.txt:1858
The compaction copy contains only live data (db/kv/backup/backup.go:120-123), not a second full copy of the bloated database. This generated paragraph therefore overstates the required filesystem headroom; update the source to say “a second copy of its live data” and regenerate this file.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Documents 3.7 changes the docs did not cover. Each claim was checked against this branch's source.
Configuration
--caplin.builder.allow-private-urls: lets validator-submitted builder URLs resolve to private or loopback addresses; does not apply to--caplin.mev-relay-url(cl/beacon/builder/client.go).--http.compression: Zstandard or gzip per the client'sAccept-Encoding, for responses of 1 KiB or more (node/rpcstack.go).MAX_REORG_DEPTH: once a finalized block is known (after the initial sync cycle), reorg data is kept back to it instead; the value still caps newPayload ancestor download and side-fork validation (execution/execfinality/context.go).RPC
erigon_getLogs,erigon_getLatestLogs,erigon_getLogsByHash: block time asblockTimestamp(wastimestamp).ValidateBlobPrerequisites).eth_syncingduring snapshot download and the neweth_subscribe("syncing").blockTimestampon filter and subscription logs.callTracerwithLog: block-wide log indices; failed CREATE frames have noto.Operations
GOMEMLIMITdefaults to 80% of the lower of RAM and the cgroup limit (FAQ Fix CircleCI #29, OOM entry).Branch-specific (
release/3.7)353024, 6 Oct 2026 13:53:36 UTC (sepolia.jsonamsterdamTime,GloasForkEpoch), and theblockTimestamprename.eth_syncingreportsstartingBlockas0x0; the subscription reports the block it first saw syncing.The Sepolia bullet names v3.7.1 as the minimum: #24386, #24441 and #24474 (Glamsterdam/Gloas) landed after v3.7.0. Both versions are declared with
llms-pinned-version.Same content on
main: #24490.npm run build, the script tests andgenerate-llms.py --checkpass. Pushed with--no-verify: the pre-push review flags the frozen v3.6 archive's Shutter page on everyrelease/3.7branch.