Skip to content

v1.1.0

Choose a tag to compare

@djh58 djh58 released this 11 Nov 15:59
· 56 commits to main since this release
v1.1.0
70f469f

v1.1.0 catches the reorgs the block sync used to miss and repairs the chain index, serves the esplora routes the API docs listed, models more of qbit's relay policy, measures broadcasts in qbit's weight (so wallets can send standard transactions with up to the node's limit of inputs), builds the Docker images as production does, fixes the SSR build, and removes the unused Lightning backend. It has a database migration (113, dropping the unused Lightning tables) and changes the Docker build: read Upgrading below before you deploy.

Added

  • GET /api/address/:address/txs/chain, with an optional /:last_seen_txid, and GET /api/address/:address/txs/mempool, which the API docs listed but the backend didn't serve. They pass through to qbit-electrs, so they need MEMPOOL.BACKEND esplora. The other backends answer 405.
  • GET /api/block/:hash/status, GET /api/tx/:txid/raw, GET /api/tx/:txid/outspend/:vout and GET /api/tx/:txid/merkleblock-proof, the other routes the API docs listed but the backend didn't serve. Block status and the merkleblock proof need MEMPOOL.BACKEND esplora, and the other backends answer 405. The raw transaction and the outspend work with every backend.
  • MEMPOOL.HTTP_HOST, and the Docker image's MEMPOOL_HTTP_HOST: the IP address the backend's HTTP port listens on. It defaults to "", every interface, as before, and the production/ templates set it to 127.0.0.1. A host name such as localhost stops the backend at startup.
  • The WebSocket blocks message sent after a reorg carries "reorg": true, so pages can follow a reorg to a shorter chain.
  • An optional github_token build secret for the frontend Docker image. The build lists the mining pool logos through the GitHub API, and a token keeps it under GitHub's rate limit.
  • CI tests a fee bump (RBF) in the browser, checks the translations for stale English text, and runs the Rust block-template builder's tests and a RustSec audit of its crates.

Fixed

  • The Block Fees, Block Fees Vs Subsidy and Block Rewards graphs draw their QBT data. They were empty because their queries needed a USD price for each block, which qbit doesn't have. GET /api/v1/mining/blocks/fees/:interval and GET /api/v1/mining/blocks/rewards/:interval now return rows, without USD.
  • Search suggests addresses when qbit-electrs runs with --address-search. In esplora mode, GET /api/address-prefix/:prefix used to answer 500 for every prefix. It now answers [] when qbit-electrs has address search off, or when the text can't begin an address.
  • A mining graph opened at a range whose button is hidden opens at ALL. Examples are pools dominance's default 6M and the dashboard's link to the hashrate chart at 1Y. Such a graph used to load that range with no button highlighted. The visitor's saved range is kept.
  • After an upgrade, transactions restored from the mempool cache are classified by the new release's rules. They used to keep the older release's flags, so after the v1.0.1 upgrade some qscriptions weren't highlighted in the block preview.
  • Error messages over HTTP/2 show the status code alone, such as 400, instead of 400 OK.
  • Enterprise mode (customize.enterprise) applies from the first page a visitor opens. It used to wait until a transaction, tracker or preview page loaded. A branding without a title keeps the default page title.
  • With ESPLORA.FALLBACK hosts, a healthy qbit-electrs host is no longer failed over because it refused bad requests, such as invalid addresses. Only the host's own errors, timeouts and errors from a proxy in front of it count, and they count against the host that failed.
  • With MEMPOOL.BACKEND electrum, address and scripthash histories page after an after_txid as they do in esplora mode. An after_txid missing from the history used to return the first page again, and the next page repeated the newest transaction. One that qbitd doesn't have gets 422, and a malformed one gets 400 on the scripthash route too.
  • The backend catches reorgs with indexing off, and a block replaced at the same height reaches pages within one poll, instead of when the next block arrives. With indexing off, a reorg used to leave the replaced block cached below the new ones.
  • At startup, cached blocks that qbitd doesn't have, as in a disk or Redis cache from before a chain wipe, are dropped and synced again from qbitd. The tip height and the block list used to show the old chain until qbitd passed it. Blocks above qbitd's tip stay cached while qbitd catches up on the same chain, or is still syncing.
  • With indexing on, chain validation makes qbitd's block the only canonical one at each height it checks, and indexes hashrates and CPFP clusters again from the first block it repairs. A chain wipe used to leave the old chain's blocks canonical below the cached ones, in block lists and mining stats.
  • On a chain of up to 8 blocks, such as a fresh regtest chain, the backend caches every block, not only genesis and the tip.
  • After a reorg to a shorter chain, such as a chain wipe, pages lower their tip and show new blocks as they arrive. They used to ignore new blocks until the chain passed the old tip, so the mempool blocks stood still and a transaction confirmed in one of them still showed as unconfirmed.
  • A failed block or mempool update, such as during a qbitd or qbit-electrs outage, no longer logs a false stalled error two minutes later.
  • /tx/preview shows a transaction's weight and vsize as qbit counts them. It overstated both by three times the size without witnesses, and judged standardness by them.
  • POST /api/v1/tx, /api/v1/tx/push and /api/v1/txs/test accept standard transactions with more than about 100 KB of non-witness data. Their size check used Bitcoin's weight.
  • With MEMPOOL.ALLOW_UNREACHABLE false, pushed P2MR spends whose leaf script holds OP_FALSE OP_IF are refused, as that setting intends. The check let every P2MR spend through.
  • Opcodes 0xbb to 0xbd show as OP_CHECKTEMPLATEVERIFY, OP_CHECKDATASIGPQC and OP_CHECKDATASIGADDPQC, as qbitd names them, where the explorer writes its own ASM: on /tx/preview, with MEMPOOL.BACKEND none, and for mempool transactions restored from the cache. They showed as OP_RETURN_187 to OP_RETURN_189.
  • With MEMPOOL.BACKEND none or electrum, OP_RETURN and other unspendable outputs show as unspent, as qbit-electrs has them. They showed as spent.
  • npm run build:ssr builds the server-side rendering (SSR) server again, on Angular 20 and Express 5, and the server renders pages. It renders for the host names in NG_ALLOWED_HOSTS, by default localhost, 127.0.0.1 and the PROD_DOMAINS with their subdomains. Any other Host, or one that isn't a host name or IP address with an optional port, gets 400. A page that takes longer than SSR_RENDER_TIMEOUT_MS, 5 seconds by default, is sent for the browser to render. CI builds the server and renders a page.

Changed

  • The non-standard flag follows more of qbit's relay policy. Dust thresholds use qbit's spend sizes and round up, so a P2MR output is dust below 2,855 bits instead of 82. TRUC transactions over 50,000 vB, outputs to witness versions 3 to 16, and P2MR spends with an annex or beyond qbit's witness limits are flagged. Taproot script-path spends no longer skip the remaining checks. Sigops are counted over decoded scripts, as qbitd counts them, instead of over ASM text.
  • /tx/preview also checks the rules qbitd applies when a transaction enters its mempool: outputs only to P2MR, OP_RETURN or the anchor, except on regtest; TRUC's version inheritance, its one unconfirmed ancestor and its 45,500 vB limit for a child; and spending all of an unconfirmed parent's dust. It loads the unconfirmed parents to do this, and says so when it can't. Elsewhere the flag still judges each transaction on its own.
  • P2MR inputs with an annex get the annex flag, as Taproot inputs do.
  • Transactions no longer show the struck-through Taproot badge or its hint to save fees with Taproot, which don't fit qbit's P2MR addresses.
  • Reward stats on the mining dashboard cover the last 1,440 blocks, a day of qbit blocks, instead of 144. Translated pages show those labels in English until their translations catch up.
  • MEMPOOL.RUST_GBT defaults to true, as it does upstream, so projected blocks come from the faster Rust block-template builder. The builder now uses qbit's weights. It used to overstate fee rates fourfold. The Docker image and the production/ templates still set it to false.
  • MEMPOOL.INDEXING_BLOCKS_AMOUNT defaults to 110,000 blocks, about 76 days of 60-second blocks, up from 11,000, in the Docker image too.
  • Malformed txids, block hashes and scripthashes get 400 instead of 501, on every route, including GET /api/block/:hash/txid/:index. GET /api/v1/transaction-times without txId[] and GET /api/v1/txs/outspends without txids get 400 instead of 500.
  • An unknown transaction or block gets 404 instead of 500 from eight more routes, such as GET /api/tx/:txid/outspends and GET /api/block-height/:height. A malformed height gets 400. With MEMPOOL.BACKEND none, GET /api/tx/:txid/merkle-proof answers 405.
  • With MEMPOOL.BACKEND none or electrum, the transaction routes answer 503 and name qbitd's txindex when it's disabled or still building. They answered 404, as for a transaction that doesn't exist.
  • A relative MEMPOOL_CONFIG_FILE is resolved against the backend's working directory, not backend/dist/. A file found only at the old location is still read, with a deprecation warning.
  • With CORE_RPC.COOKIE true, the backend authenticates only with qbitd's RPC cookie, even when COOKIE_PATH is empty. It no longer falls back to USERNAME and PASSWORD. It logs an error at startup when it can't read the cookie. When qbitd restarts with a new cookie and turns a call away, the backend retries that call once with the new cookie. SECOND_CORE_RPC works the same way.
  • The production/ templates read qbitd's RPC cookie at /var/lib/qbit/.cookie instead of taking an RPC user name and password.
  • In the Docker image, nginx looks the backend's host name up when a request comes in, through DNS only, so recreating the api container no longer needs a web restart. A name that only /etc/hosts knows, or a Kubernetes short name, needs an IP address or a full name instead.
  • nginx-mempool.conf keeps the backend's address in $mempool_backend, and passes /api/v1 paths to the backend as the client sent them.
  • A static file that doesn't exist, such as a script name from another build, answers a 404 that isn't cached, instead of the app page, so a CDN or browser never stores HTML under a script's name. Files that exist and app routes are unchanged.
  • Building the Docker images needs BuildKit, Docker's default builder since Docker 23. The Dockerfiles pin their base images by digest, and the backend image checks NodeSource's setup script against a SHA-256 before running it.
  • rust/gbt/rust-toolchain is in TOML form, which Dependabot can read. The backend's install stops when the file names no toolchain channel.
  • The site header links to qbit.org's new Mine page.
  • The frontend lists express 5 and http-proxy-middleware 2, which the SSR server needs, as runtime dependencies.
  • Backend dependencies: bitcoinjs-lib 6.1.3 to 6.1.8 and bip174 2.1.0 to 2.1.1, for a critical advisory in code the explorer doesn't call, and the test and lint tools' handlebars, brace-expansion, browserslist and js-yaml, for theirs.
  • Rust block-template builder dependencies: bytes 1.9.0 to 1.12.1, tokio 1.43.0 to 1.53.3, and tracing-subscriber 0.3.19 to 0.3.20, for four RustSec advisories.

Removed

  • The unused Lightning backend: its /api/v1/lightning/ routes, the LIGHTNING, LND, CLIGHTNING and MAXMIND settings and their Docker variables, and its seven database tables, which the backend drops at startup (schema version 113). A config file that still has those settings loads, with a startup warning. /api/v1/backend-info keeps lightning, now always false. The Docker image no longer bundles MaxMind's GeoLite2 databases.

Documentation

  • The API docs' examples use live qbit mainnet and testnet4 blocks, transactions and mining pools, with their real responses where the explorer has them. The WebSocket address-tracking examples subscribe to qbit addresses.
  • The WebSocket track-address, track-addresses and track-mempool examples answer with messages captured on qbit mainnet, and on testnet4 with a testnet4 payment. The block status and merkleblock proof examples use qbit blocks, and the docs say that the proof carries no AuxPoW data.
  • production/README.md and docker/README.md show how to authenticate the backend and qbit-electrs with qbitd's RPC cookie, and how to get address suggestions from qbit-electrs. docker/README.md also says how nginx looks up the backend's host name, and which names need an IP address.
  • frontend/README.md shows how to build and run the SSR server, and how to set its allowed hosts and render timeout.
  • CONTRIBUTING.md says how a pull request lands: a maintainer squashes it into one signed commit and merges it with a merge commit. The branch rulesets are in .github/rulesets/. It also says how to fix a translation that CI reports as stale.

Upgrading

Database

  • Migration 113 drops the Lightning backend's seven tables at the first start: nodes_records, nodes_sockets, nodes, node_stats, channels, lightning_stats and geo_names. The schema version goes from 112 to 113, and an existing database logs one notice naming the tables it dropped. The drop can't be undone, but qbit never ran Lightning, so they hold no qbit data. v1.0.1 still starts on a schema-113 database, as long as Lightning stays off, as it is by default.
  • Without MEMPOOL.INDEXING_BLOCKS_AMOUNT in your config, or MEMPOOL_INDEXING_BLOCKS_AMOUNT in Docker, the indexing window goes from 11,000 to 110,000 blocks. The first start after the upgrade backfills about 99,000 older blocks, or every block on a chain shorter than 110,000. Expect a longer first start, about six qbitd or qbit-electrs calls per block, and about 99,000 more blocks rows. Set it to 11000 to keep the old window, or to -1 to index every block. The production/ templates set -1 and don't change.
  • With indexing on, chain validation now makes qbitd's block the only canonical row at each height it checks. If an earlier release left another chain's rows canonical, for example after a chain wipe, the first validation after the upgrade repairs them. It deletes the hashrates from a week before the first repaired block and, with MEMPOOL.CPFP_INDEXING on, the CPFP clusters from that height up, and indexes them again. A database without such rows sees no change. Rows above qbitd's tip stay until its chain passes them, and the lowest indexed height isn't checked.
  • At the first start, the backend checks the block cache it restores from disk or Redis against qbitd. Blocks qbitd has replaced are dropped and synced again; with indexing on, their rows are marked stale and their hashrates and CPFP clusters indexed again. Blocks above qbitd's tip are kept while the cache's highest block at or below that tip is on qbitd's chain, since qbitd may only be catching up. A cache with no block at or below qbitd's tip is kept while qbitd reports itself syncing (initial block download, or more headers than blocks), and dropped once it reports itself synced. So a qbitd restarted after a long stop, which can report itself synced before it catches up, costs a re-sync of those blocks.
  • Block summaries already stored keep the flags they were saved with. The qscription fix and the new non-standard rules apply to blocks mined after the upgrade, and nothing reclassifies older ones.
  • If you ran with Lightning on, you can delete ln-funding-txs-cache.json from the cache directory.

Settings in mempool-config.json

  • New: MEMPOOL.HTTP_HOST defaults to "", which listens on every interface, as before. It takes an IP address, such as 127.0.0.1 or ::1. Anything else, localhost included, stops the backend at startup with an error. It applies to the TCP port only: a MEMPOOL.UNIX_SOCKET_PATH listener is separate. The production/ templates set 127.0.0.1. If you deploy from them and anything but a local nginx reaches port 8999, change it.
  • Removed: the LIGHTNING, LND, CLIGHTNING and MAXMIND sections. A file that still has any of them loads, the backend ignores them, and it prints Ignoring config file sections that only the removed Lightning backend read: ... at startup. Older copies of mempool-config.sample.json had all four; delete them to silence the warning.
  • Default changed: MEMPOOL.RUST_GBT goes from false to true. Without the key, projected blocks come from the Rust builder in rust/gbt, which now uses qbit's weights. The Docker image (MEMPOOL_RUST_GBT) and the production/ templates still set false. The startup warning about RUST_GBT is gone.
  • Default changed: MEMPOOL.INDEXING_BLOCKS_AMOUNT, under Database above.
  • A relative MEMPOOL_CONFIG_FILE is resolved against the working directory, which is backend/ under the npm scripts, instead of backend/dist/. If the file exists only at the old location, the backend still reads it and warns that the fallback will be removed. If both exist, it reads the working-directory file and warns, naming both. Absolute paths and the unset default (backend/mempool-config.json) work as before.
  • With CORE_RPC.COOKIE or SECOND_CORE_RPC.COOKIE true, the backend no longer falls back to USERNAME and PASSWORD, even when COOKIE_PATH is empty. If you relied on that, set COOKIE_PATH to qbitd's cookie, or set COOKIE to false. Until you do, RPC calls fail, and the backend logs an error at startup that names the setting and the fix. When qbitd restarts with a new cookie and turns a call away, the backend retries the call once with the new cookie, so the backend doesn't need a restart.
  • With MEMPOOL.ALLOW_UNREACHABLE false (the default is true), POST /api/v1/tx, /api/v1/tx/push and /api/v1/txs/test now refuse a P2MR spend whose leaf script holds OP_FALSE OP_IF, with code -5. Earlier releases let every P2MR spend through.

production/ templates and bare-metal nginx

  • The templates authenticate with qbitd's RPC cookie ("COOKIE": true, "COOKIE_PATH": "/var/lib/qbit/.cookie"). They no longer have the __QBIT_RPC_USER__ and __QBIT_RPC_PASS__ placeholders, so tooling that fills those in now does nothing. Add the backend's user to the qbit group (qbitd needs rpccookieperms=group) and log in again. Or keep the password: set COOKIE to false, then set USERNAME and PASSWORD.
  • nginx-mempool.conf keeps the backend's address in set $mempool_backend "127.0.0.1:8999"; near its top, and the API and WebSocket locations proxy to it. If you changed the address in each proxy_pass, change it there instead, and keep an IP address. The resolver 127.0.0.11 line next to it is for the Docker image and has no effect on an IP address. If your server block has a resolver of its own, delete this one: nginx refuses two in one block.
  • Under /api/v1, nginx now passes the path to the backend as the client sent it. It no longer decodes %2F or collapses // and .. there. The other proxied locations still pass a normalized path.

Docker

  • Building the images needs BuildKit, Docker's default builder since Docker 23. docker buildx build, which docker/README.md uses, always uses it.
  • The Dockerfiles pin their base images by digest. If NodeSource republishes setup_26.x, the backend build stops at its SHA-256 check until both stages of docker/backend/Dockerfile get the new checksum (docker/README.md).
  • The frontend build lists the mining pool logos through the GitHub API, which allows 60 unauthenticated requests an hour per IP address. To raise that limit, set GITHUB_TOKEN and add --secret id=github_token,env=GITHUB_TOKEN to the frontend build. The token isn't stored in the image.
  • New variable MEMPOOL_HTTP_HOST (MEMPOOL.HTTP_HOST), empty by default. Under Compose, leave it empty: the web container reaches the backend over the Compose network.
  • MEMPOOL_INDEXING_BLOCKS_AMOUNT defaults to 110,000 (see Database). Set it to 11000 to keep the old window.
  • The LIGHTNING_*, LND_*, CLIGHTNING_SOCKET and MAXMIND_* variables are gone. start.sh ignores them if they're still set.
  • The backend image has no /backend/GeoIP and no MaxMind data, and docker/init.sh no longer downloads them.
  • web: nginx looks BACKEND_MAINNET_HTTP_HOST up when a request comes in and keeps the answer for 10 seconds, so a recreated api container no longer needs a web restart. nginx starts even when the name doesn't resolve yet, and answers 502 for the API until it does; it used to exit. At every start, the entrypoint sets nginx's resolver to the first usable nameserver in the container's /etc/resolv.conf: 127.0.0.11 on a user-defined network such as Compose's, and another server in Kubernetes, in Podman or on the default bridge network, which it logs.
  • nginx reads neither /etc/hosts nor search domains. A name that only /etc/hosts knows (extra_hosts, --add-host, --link), a Kubernetes short name (use api.<namespace>.svc.cluster.local), and an IPv6-only name behind an IPv4 nameserver no longer resolve. Use an IP address for those, with an IPv6 address in brackets. See "Backend host names" in docker/README.md.
  • CORE_RPC_COOKIE_PATH still defaults to empty, so with CORE_RPC_COOKIE set to true, set it too (see the cookie note above). docker/docker-compose.yml still uses an RPC password; docker/README.md shows the cookie setup.

Building from source

  • Build the backend as before, with npm ci and then npm run build (production/README.md). npm ci recompiles the rust/gbt addon, which this release changes.
  • rust/gbt/rust-toolchain is now TOML ([toolchain], channel = "1.84"), which rustup reads as it read the one-line form. Read the channel with rust/gbt/scripts/toolchain-channel.sh, not cat. The backend's install stops when the file names no channel.
  • The frontend's package.json lists express 5 and http-proxy-middleware 2 as dependencies, so npm ci --omit=dev installs them.

Frontend

  • With customize.enterprise set, enterprise mode applies from the first page a visitor opens. It sets the title from branding.title, turns off the accelerator, and clears the testnet, testnet4, signet and regtest flags and explorer URLs. On a non-mainnet instance, that turns off the instance's own network from the first page, so set enterprise on mainnet instances only, as the docs say.
  • /tx/preview makes more API requests: /api/v1/prevouts for a PSBT's own prevouts, and /api/tx/:txid for each unconfirmed parent.
  • SSR server: npm run build:ssr and npm run serve:ssr work again. The server renders only for the host names in NG_ALLOWED_HOSTS, Angular's environment variable, read at startup: comma-separated, with *.example.org for subdomains. Without it, it renders for localhost, 127.0.0.1 and the PROD_DOMAINS with their subdomains. Any other Host, or one that isn't a host name or IP address with an optional port, gets 400. If your proxy passes the visitor's Host (proxy_set_header Host $host; in nginx), set NG_ALLOWED_HOSTS or PROD_DOMAINS to your site's host names.
  • SSR server: a page that doesn't render within SSR_RENDER_TIMEOUT_MS (read at startup, 5000 by default) is sent unrendered for the browser to render, with Cache-Control: no-store, and its path is logged. The render goes on in the background.

qbit-electrs

  • For address suggestions in search, run qbit-electrs with --address-search, and rebuild an existing index from an empty --db-dir. Without the flag, search suggests no addresses, and GET /api/address-prefix/:prefix answers [] where it used to answer 500.

API changes clients may see

  • A malformed txid, block hash or scripthash gets 400 instead of 501, on every route that checks one, GET /api/block/:hash/txid/:index included; the v1.0.1 notes listed that route as answering 501. GET /api/v1/transaction-times without txId[] and GET /api/v1/txs/outspends without txids get 400 instead of 500.
  • GET /api/address/:address/txs/chain[/:last_seen_txid] and GET /api/address/:address/txs/mempool used to answer 404. In esplora mode they answer 200, and under the none and electrum backends 405. An invalid address or a malformed last_seen_txid gets 400, and a history qbit-electrs refuses as too large gets 413.
  • Under the electrum backend, an after_txid outside an address or scripthash history gets an empty page, and one that qbitd definitely doesn't have gets 422. Both used to get the first page again. With qbitd's txindex off or still building, the second case is a 500. GET /api/scripthash/:scripthash/txs answers 400 for a malformed after_txid.
  • GET /api/address-prefix/:prefix answers 200 with [] instead of 500, both when qbit-electrs has address search off and for text that can't begin an address.
  • GET /api/v1/mining/blocks/fees/:interval, GET /api/v1/mining/blocks/fees?from=&to= and GET /api/v1/mining/blocks/rewards/:interval return rows instead of []. The rows have no USD field.
  • POST /api/v1/tx, /api/v1/tx/push and /api/v1/txs/test compare qbit's weight, the serialized size, with MAX_PUSH_TX_SIZE_WEIGHT, so standard transactions with more than about 100 KB of non-witness data are no longer refused as too large (code -3).
  • With MEMPOOL.BACKEND none, and for mempool transactions restored from the cache, ASM fields name opcodes 0xbb to 0xbd OP_CHECKTEMPLATEVERIFY, OP_CHECKDATASIGPQC and OP_CHECKDATASIGADDPQC, as qbit-electrs does, instead of OP_RETURN_187 to OP_RETURN_189.
  • The WebSocket blocks message sent after a reorg carries "reorg": true. The lists sent on connect, on a first want and on refresh-blocks don't. Clients that don't know the field can ignore it.
  • After a chain wipe, GET /api/blocks/tip/height answers qbitd's height once the backend has dropped the old cached blocks (see Database), not the cached height.
  • The 24 routes under /api/v1/lightning/ are gone. With Lightning off, the default, they already answered 404.
  • /api/v1/backend-info and the WebSocket backendInfo message keep their lightning field, now always false.
  • New routes: GET /api/block/:hash/status and GET /api/tx/:txid/merkleblock-proof (text/plain) under the esplora backend, 405 under none and electrum; GET /api/tx/:txid/raw (application/octet-stream) and GET /api/tx/:txid/outspend/:vout under every backend. A malformed block hash, txid or output index gets 400. A merkleblock proof of an unknown or unconfirmed transaction gets 404. Under esplora, block status and outspend pass qbit-electrs's answers through: an unknown block gets in_best_chain: false, and an output with no spend, whether it exists or not, gets spent: false. Under none and electrum, outspend answers as tx/:txid/outspends does, and 404 for an unknown transaction.
  • An unknown transaction or block gets 404 instead of 500 from GET /api/tx/:txid/outspends, /api/tx/:txid/merkle-proof, /api/block/:hash/header, /txids, /txs and /txs/:index, and /api/block-height/:height; under the none and electrum backends, from /api/block/:hash and /raw too. /api/block-height/:height answers 400 for a height that isn't digits, where it answered 500, and 404 above 2,147,483,647. Other failures stay 500.
  • Under the none backend, GET /api/tx/:txid/merkle-proof answers 405 instead of a 500 that repeated the internal error.
  • Under the none and electrum backends, every transaction route, GET /api/tx/:txid and /status included, answers 503 when qbitd's txindex is disabled or still building, with a message that says so. It answered 404. qbitd's "No such mempool or blockchain transaction", from a synced txindex, still gets 404.
  • Under the none and electrum backends, GET /api/tx/:txid/outspends and /outspend/:vout answer "spent": false for OP_RETURN outputs and scripts over 10,000 bytes, which they reported as spent.
  • The SSR server answers 400 for a host outside its list and for a malformed Host header.

Logs

  • New at startup, when they apply: the ignored-sections warning, the MEMPOOL_CONFIG_FILE fallback warning, the RPC cookie errors, and Mempool Server is running on <host> port <port> with HTTP_HOST set. Migration 113 logs Lightning tables have been dropped: ... once.
  • Cached blocks #<a> to #<b> are not on the node's best chain (tip #<n>), dropping them from the cache (warning) when the backend rolls its cache back, and Chain divergence detected at block <h>: repairing ... (warning) when validation repairs the database. Waiting for a qbitd that is behind logs at debug level.
  • A failed block or mempool update no longer logs a false $updateBlocks stalled or $updateMempool stalled error two minutes later. A run that hangs still does.
  • When qbit-electrs refuses a bad request (an invalid address, its UTXO and history limits, or an unknown after_txid), the backend logs one debug line instead of two warnings, and the refusal no longer counts toward failing a host over.
  • Docker web logs one line at start when nginx resolves through a nameserver other than 127.0.0.11, or when /etc/resolv.conf has none it can use.
  • The SSR server logs its allowed hosts at startup. Angular logs an error line for each request from another host, and a page that hits the render timeout logs its path.

Full changelog: v1.0.1...v1.1.0