Repository navigation
Releases
Releases · ZcashFoundation/zinder
v0.15.0
Immutable
release. Only release title and notes can be modified.
[0.15.0] - 2026-11-15
Changed
- Releases now prove their binaries before publishing anything. Both builds of each platform start as soon as the tag is validated, and every release image is packaged from those binaries before the release is approved. Public SDK crates, container images, and the GitHub Release are published only after both builds match. A build, reproduction, or packaging failure no longer leaves a published crate version without matching binaries. (#206)
- Explorer views now keep per-height undo rows only for the reorg window, and ingest deletes older undo rows from an existing view store when it opens it, without a rebuild. New transparent-address delta and activity rows store each transaction id as 32 bytes instead of 64 hex characters, while rows already written keep reading. On a transparent-heavy fixture the two changes cut view storage by 41%. Rolling back to an earlier release requires rebuilding the Explorer views, because earlier releases read the rows this release writes with an empty transaction id. (#208)
Fixed
- A transparent-address ranking build that trails Explorer views still rebuilding from canonical history now catches up and joins them, instead of falling further behind until replay reaches the tip, so address detail and activity return far sooner on read-limited storage. View replay waits for such a build while
zinder_materialized_view_replay_waiting_for_ranking_buildis1, and stops waiting once the build, outside pauses it shares with replay, neither applies a page nor warms its cache further for 5 minutes. Spend resolution also issues fewer device reads, the prevout resolution metrics gain acachelabel that names the walker, andzinder-ingestexports its canonical reader's RocksDB gauges under thecanonical_secondarystore role. (#209)
v0.14.0
Immutable
release. Only release title and notes can be modified.
[0.14.0] - 2026-11-15
Fixed
- Explorer view replay and the transparent-address ranking build no longer slow to a few dozen blocks per minute when a deployment caps glibc malloc arenas, as
MALLOC_ARENA_MAX=2does; on a Mainnet-shaped fixture with that setting, replay ran about 24 times faster. Every service binary now allocates through jemalloc on Linux, including RocksDB's native allocations, so glibc malloc variables such asMALLOC_ARENA_MAXno longer apply. Remove them from the deployment and tune the allocator withMALLOC_CONFinstead. Peak memory stayed within about 10% of glibc's on the same workload. (#199) ExplorerQuery.TransactionDetailno longer fails withARTIFACT_UNAVAILABLEfor mined transactions with transparent outputs while the Explorer's transparent-outpoint spend view does not cover the response's chain epoch, such as during view replay or while the view and the wallet reader are a block apart at the tip. The response carries the transaction's facts, location, inputs, and outputs, omits everyspent_by, and liststransparent_outputs.spent_byinfreshness.unavailablewithUNAVAILABLE_STALE; clients must treat an absentspent_byas unknown, not unspent, when that entry is present. The Cipherscan adapter reports that output state asspent: null. Thezinder-wallet-serving-explorerimage now startszinder-exploreras soon as wallet serving answers whenZINDER_INGEST__EXPLORER_VIEWSis false, instead of after a 10-minute wait for view admission.zinder-queryandzinder-compat-lightwalletdlogwallet_serving_pair_publisher_refresh_failedat debug level, not as a warning, when a refresh only found its secondaries an epoch apart within the replica lag threshold. (#197)
v0.13.0
Immutable
release. Only release title and notes can be modified.
[0.13.0] - 2026-11-15
Added
- Releases now publish the Explorer. The
ghcr.io/zcashfoundation/zinder-explorerimage can replace the locally built Explorer in the Compose overlay when it is pulled and started without a build, and the GNU/Linux binary archives include thezinder-explorerexecutable. For hosts that attach one volume to one container, such as Railway, the newghcr.io/zcashfoundation/zinder-wallet-serving-explorerimage runs wallet serving with Explorer views enabled on ingest. It servesExplorerQueryon port 9068 and Explorer operations on port 9069, and starts the Explorer once wallet serving answers. If the Explorer exits, only the Explorer restarts, and wallet serving continues. The image healthcheck covers wallet serving alone, so monitor Explorer readiness on port 9069.ZINDER_EXPLORER__*variables reach only the Explorer. A volume built byzinder-wallet-servingcan switch to this image: ingest then builds the Explorer views from canonical history. See the wallet-serving deployment runbook. (#165)
Fixed
zinder-projectorno longer exits withcanonical secondary did not reach the writer fence before the convergence deadlinewhen a burst of blocks keeps the canonical writer committing ahead of it, which restarted every process in the single-volume wallet-serving runtime. It keeps applying the blocks its canonical secondary already holds while the writer is ahead, including right after a restart, each still authenticated against the writer's retained events, and when an attempt to reach the writer lastsprojector.fence_convergence_timeout_secondsit logscanonical_secondary_behind_writer, reportsreplica_lagging, renews its canonical retention lease, and tries again. A ready projector also staysreadywhile its wallet trails the writer by any number of blocks, as long as it applies a block at least every 30 seconds and the trail lasts less than 2 minutes; it previously reportedsyncingwhenever it trailed by two or more blocks. (#192)- Slow or abandoned
TransparentAddressUnspentOutputsreads no longer stopzinder-queryorzinder-compat-lightwalletdfrom publishing new chain epochs for every client. The read now sends outputs as it reads them, 1,024 at a time, reads storage only as fast as the client consumes, and stops when the client cancels or disconnects; the first outputs of a large address arrive after one page instead of after the whole set. A request may keep reading a serving pair for at most 30 seconds after a newer one replaces it and leaves retention. An unspent-output or full-block stream still reading past that limit ends withFAILED_PRECONDITIONandCHAIN_EPOCH_PIN_UNAVAILABLE, which clients already treat as a signal to restart at a newer epoch.zinder_wallet_serving_pair_publisher_revoked_pairs_totaland thewallet_serving_pair_reads_revokedwarning report each such revocation. (#193) - Explorer view replay that resumes partway through history, as after a restart or upgrade, now warms its prevout cache from canonical history before it continues, instead of resolving most transparent spends through one canonical read at a time, and the canonical reads that remain run on up to 16 threads; the transparent-address ranking build does the same, and both log
materialized_view_prevout_cache_warmed. On a store whose Explorer views were built without the transparent-address ranking, ingest now builds the ranking beside the view replay and joins it at the height the other views have reached, soexplorer.transparent_address.activity_v2andexplorer.transparent_address.ranking_v1are advertised long before replay reaches the tip; ingest logstransparent_address_ranking_joined_replay, andzinder_materialized_view_transparent_address_ranking_build_heightandzinder_materialized_view_transparent_address_ranking_build_target_heightreport the build's progress.zinder_materialized_view_prevout_cache_outputs{cache}andzinder_materialized_view_prevout_cache_evicted_outputs_total{cache}report prevout cache use. Explorer view files now compress with ZSTD, which used about a third less disk than the previous Snappy default on a transparent-dense fixture; existing view stores open in place and shrink as compaction rewrites their files. (#194)
v0.12.0
Immutable
release. Only release title and notes can be modified.
[0.12.0] - 2026-11-14
Fixed
zinder-projectorno longer reportssyncingwithlag_blocks: 1for a moment after every block: a ready projector within one block of the canonical tip staysreadywhile it applies the block, and reportssyncingonly if 30 seconds pass without it applying a block or it trails by more than one block.zinder-querynow exits promptly onSIGTERMeven while wallets holdChainEventsorMempoolEventsstreams open: it ends every open stream withUNAVAILABLEand readiness causeshutting_down, and stops waiting for in-flight calls after 10 seconds, so it stops within the wallet-serving supervisor's 20-second grace window. (#182)ExplorerQuery.TransparentAddressActivityno longer fails withFAILED_PRECONDITIONwhile the transparent-address ranking is ahead of the requested chain epoch, which happened for most of every block interval. Address summaries, ranking coverage, and activity rows now describe the epoch's visible tip. A read still fails withFAILED_PRECONDITIONwhen the ranking holds a different block than the epoch, including while it trails the epoch on a replaced branch, or when it would need to rewind more than 1,000 blocks. The Cipherscan adapter no longer marks/api/address/:addressdegraded only because the Explorer views lead the pinned epoch; it does so when ranking coverage stops short of the epoch's tip. (#183)- Explorer view replay and the transparent-address ranking build no longer re-read the producing block for each transparent spend of an output they already replayed, which made transparent-heavy history such as early Mainnet replay at a few hundred blocks per minute; they keep up to about 375 MB of unspent outputs in memory while they walk history and release it at the tip or under memory pressure. Explorer view replay, the ranking build, and historical view backfills now also pause while the volume holding
storage.pathhas less available space than the newingest.explorer_views_min_free_bytes(ZINDER_INGEST__EXPLORER_VIEWS_MIN_FREE_BYTES, default 10 GiB,0disables), loggingmaterialized_view_replay_volume_space_low, and resume on their own once a tenth more is available; canonical ingest and wallet serving keep running. Existing Explorer view stores resume in place. (#184)
v0.11.1
Immutable
release. Only release title and notes can be modified.
[0.11.1] - 2026-11-14
Fixed
- The
zinder-canonical-runtimeandzinder-wallet-servingimages now carry/usr/local/lib/zinderas a traversable directory. In 0.11.0 the image build gave that directory mode 0644, which stopped the release from publishing images, binary archives, and a GitHub Release for that version. Zinder 0.11.0 exists only as thezinder-core,zinder-proto, andzinder-clientcrates; deploy 0.11.1 instead, which contains every 0.11.0 change. (#174)
v0.10.0
Immutable
release. Only release title and notes can be modified.
[0.10.0] - 2026-11-13
Added
- Native
WalletQuery.TransparentOutpointSpendStatusreports, for 1 to 1,024 transparent outpoints read at one chain epoch, whether each output is unspent, spent (with the spending transaction, input index, and block), or not found. It replacesTransparentSpendsByOutpointandTransparentUnspentOutputsByOutpointwith one call, is served from the wallet projection without the Explorer views, and is advertised aswallet.read.transparent_outpoint_spend_status_v1. Inzinder-client, callChainSnapshot::transparent_outpoint_spend_status, which splits larger batches at one pinned epoch. A request with no outpoints or more than 1,024 fails withINVALID_ARGUMENTand the newTRANSPARENT_OUTPOINT_COUNT_OUT_OF_RANGEreason. On a store whose history starts after a checkpoint, a request that misses any outpoint fails withFAILED_PRECONDITIONandCANONICAL_HISTORY_UNAVAILABLEinstead of reporting it not found. The request reveals the outpoints a wallet tracks and, through them, the addresses that own them. (#143) - Broadcast results now distinguish transactions rejected for an unsupported transaction version (
UnsupportedTransactionVersion, wire value 6) from transactions that exceed the per-block action limit (ExceedsBlockActionLimit, wire value 7), where both previously appeared asUnknown.zinder-clientnow decodes unspecified and unrecognized rejection reasons asUnknowninstead of failing withMalformedResponse. Earlierzinder-clientreleases reject the new values, so release the client and update consumer pins before or together with the server. (#134) - Ingest now decodes blocks after NU7 activates (branch ID 0x77190ad9). Zinder 0.9.0 and earlier stop at the first NU7 block, so upgrade Zinder before the node serves NU7 blocks. (#136)
Changed
- Native
WalletQuery.CompactBlocksInRangenow sends the chain view once, as the first stream message, followed by bare compact blocks, instead of repeating the full chain view on every block.CompactBlocksInRangeChunkis now aoneof bodywithheaderanditem.zinder-clientkeeps the samecompact_blocks_in_rangeAPI and rejects a stream whose header is missing, repeated, or out of order withMalformedResponse. Upgrade notes: the native contract revision is now 7. Native query and ingest must run the same revision, so upgradezinder-ingestandzinder-querytogether:zinder-queryadmits only an ingest at its own revision and fails at startup against a revision 6zinder-ingest. No upgrade order is safe for native clients: a client built before this release fails on the firstCompactBlocksInRangemessage from a new server, and thiszinder-clientrefuses servers below revision 7. Release consumers such as Zally-based wallets on thiszinder-client, then cut the Zinder services and those consumers over together. Raw gRPC consumers must read the leading header and take blocks frombody.item. (#142) - Document the
GetLightdInfoandGetLatestBlockidentity contracts for lightwalletd compatibility, and compare tip-dependent fields in live parity tests only at equal tips. (#135) - Raw transaction bytes with trailing data now fail with RawTransactionParseFailed instead of being accepted for v1 to v4 transactions. Cipherscan avgBlockTimeSecs, the mining blockTime fallback, and the network-activity fallback interval now follow the target spacing at the block height (25 s from NU7) instead of a fixed 75 s. (#136)
- Wallet build cold validation reads the flushed store back once per family and compares it with the construction digest instead of re-sorting the address families, which shortens large builds. The wallet build report replaces the two cold-validation sort durations with per-family scan durations, and the storage benchmark report format moves to version 4. (#138)
Removed
- Native
WalletQuerynow serves only the transparent reads its consumers need:TransparentAddressUnspentOutputs, ascendingTransparentAddressTxIdsInRange, andTransparentOutpointSpendStatus. Removed, together with their capabilities andzinder-clientmethods:TransparentAddressBalance,TransparentOutputsByOutpoint,TransparentSpendsByOutpointandTransparentUnspentOutputsByOutpoint(replaced byTransparentOutpointSpendStatus), the three transparent mempool lookups and theirIngestControlmethods,TransparentUtxoSetSummary,ExplorerQuery.UtxoSetSummary, and thedescendinghistory flag. The lightwalletd-compatibleGetTaddressBalanceis unchanged. Wallet and ingestServerInfoand readiness no longer reportmaterialized_view_preset; Explorer view identities are reported byExplorerServerInfo. Thezinder_materialized_view_workload_infometric is removed andzinder_materialized_view_identity_infoloses itspresetlabel. Upgrade notes: the native contract revision is now 8. Native query and ingest must run the same revision, so upgradezinder-ingestandzinder-querytogether:zinder-queryadmits only an ingest at its own revision and fails at startup against a revision 7zinder-ingest. No upgrade order is safe for native clients: a client built before this release receivesUNIMPLEMENTEDfrom a removed method on a new server, and thiszinder-clientrefuses servers below revision 8. Release consumers such as Zally-based wallets on thiszinder-client, then cut the Zinder services and those consumers over together. Explorer view stores built by earlier releases are refused at startup and must be rebuilt on a fresh materialized-view path (see the deploying-on-a-VM runbook). A client built before this release that sendsdescending = truenow receives ascending results. Zallet's Zinder backend uses one removed item, thedescendinghistory flag: deletedescending: falseinbackends/zinder/src/chain.rsand the test assertion on it. Its otherzinder-clientandzinder-coreimports still exist. That was checked against the API surface, not by compiling Zallet, and upgrading it from=0.6.0also crosses every other breaking change since that release. (#143)
Fixed
- Native
WalletQuery.Transactionandzinder-clientnow report a transaction as not found only when the absence is proven, so a wallet can safely treatTxStatus::NotFoundas "never mined at this epoch". A proven absence is gRPCNOT_FOUNDwith the new reasonTRANSACTION_NOT_FOUND, whose metadata names the epoch it covers, the covered heights, and whether the live mempool was searched. A pinned lookup (at_epoch_id, or aChainSnapshot) covers that epoch's canonical chain only; an unpinned lookup also covers the writer's mempool at the same epoch. Storage faults, a missing or unserved epoch, and a block outside the best chain are now errors rather thanNotFound. A store whose history starts after a checkpoint cannot prove an absence: it answers a miss with the newCANONICAL_HISTORY_UNAVAILABLEreason (FAILED_PRECONDITION, reported asIndexerError::CanonicalHistoryUnavailablewith the first retained height), while an unpinned lookup still reports a pending transaction as in the mempool. On such a store, lightwalletd-compatibleGetTransactionnow fails a miss withFAILED_PRECONDITIONinstead ofNOT_FOUND, andis_in_mempoolfails withCanonicalHistoryUnavailableinstead of returningfalsefor a transaction that is neither pending nor mined above the floor. When the node reports a transaction mined in any block other than the tip the live mempool generation serves, ingest now ends that generation instead of removing the transaction, so live mempool reads returnUNAVAILABLEuntil the next snapshot is certified rather than briefly reporting the transaction as absent from both the chain and the mempool. The mempool half of an unpinned absence remains best effort around reorgs; use a pinned lookup (at_epoch_idor aChainSnapshot) for payment decisions. Upgrade notes: the native contract revision is now 6, and native query and ingest must run the same revision. The new client reports aNOT_FOUNDfrom an older server asIndexerError::TransactionAbsenceUnverifiedinstead ofNotFound, so upgrade servers before relying on absences; an older client keeps reading everyNOT_FOUNDasNotFound. (#105) - The release verification commands in the deployment and release runbooks now name the
ZcashFoundation/zinderrepository. Releases from v0.8.0 onward are signed byhttps://github.com/ZcashFoundation/zinder/.github/workflows/release.ymland publish images underghcr.io/zcashfoundation/, so the previously documentedgustavovalverdeidentity and image path rejected genuine releases. (#126) - Compact-blocks
GetBlockRangenow omits transactions that the requested pool types leave with no data, such as a transparent-only coinbase under the default shielded selection, so range responses match lightwalletd. Remaining transactions keep theirCompactTx.index. (#137) - RemoteChainIndex::mempool_events now fails the stream with MalformedResponse, before yielding the event, when a mempool event repeats or decreases event_sequence, reuses a cursor, or carries an empty cursor. Cursor retention is bounded to 65,536 cursors and 8 MiB per stream, matching chain events. (#141)
- The lightwalletd-compatible service now matches lightwalletd ...
v0.9.0
Immutable
release. Only release title and notes can be modified.
[0.9.0] - 2026-11-12
Added
- Each release now publishes two more multi-architecture images for hosts that attach one volume to one container.
ghcr.io/zcashfoundation/zinder-wallet-servingruns ingest, projector, and native query as supervised processes and omits lightwalletd compatibility.ghcr.io/zcashfoundation/zinder-canonical-runtimeis the ingest-only diagnostic or canary runtime. Both carry the same provenance, SBOM, and signature evidence as the four runtime images, and neither is the complete supported topology. (#128)
Removed
deploy/Dockerfile.wallet-serving,deploy/Dockerfile.railway-nocache,railway.toml,deploy/railway.canonical-runtime.toml, anddeploy/railway.wallet-serving-runtime.tomlare removed, along with theRAILWAY_DOCKER_TARGET_STAGEbuild argument. Pull the publishedzinder-wallet-servingorzinder-canonical-runtimeimage instead, or build the target of that name fromdeploy/Dockerfile. Railway services must switch from a repository build to a Docker image source, then set the healthcheck path, healthcheck timeout, restart policy, and drain window in the service settings as the deployment runbooks describe. (#128)
v0.9.0-rc.1
Immutable
release. Only release title and notes can be modified.
[0.9.0-rc.1] - 2026-11-12
Added
- Each release now publishes two more multi-architecture images for hosts that attach one volume to one container.
ghcr.io/zcashfoundation/zinder-wallet-servingruns ingest, projector, and native query as supervised processes and omits lightwalletd compatibility.ghcr.io/zcashfoundation/zinder-canonical-runtimeis the ingest-only diagnostic or canary runtime. Both carry the same provenance, SBOM, and signature evidence as the four runtime images, and neither is the complete supported topology. (#128)
Removed
deploy/Dockerfile.wallet-serving,deploy/Dockerfile.railway-nocache,railway.toml,deploy/railway.canonical-runtime.toml, anddeploy/railway.wallet-serving-runtime.tomlare removed, along with theRAILWAY_DOCKER_TARGET_STAGEbuild argument. Pull the publishedzinder-wallet-servingorzinder-canonical-runtimeimage instead, or build the target of that name fromdeploy/Dockerfile. Railway services must switch from a repository build to a Docker image source, then set the healthcheck path, healthcheck timeout, restart policy, and drain window in the service settings as the deployment runbooks describe. (#128)
v0.8.0
Immutable
release. Only release title and notes can be modified.
[0.8.0] - 2026-11-11
Removed
- Ingest no longer accepts the configuration keys of its removed batch catch-up and tip-follow paths. Remove
ingest.construction.canonical_batch_max_blocks,canonical_batch_max_artifact_bytes,canonical_batch_max_estimated_write_bytes,canonical_batch_min_blocks_before_estimated_write_close,commit_reassembly_max_queued_artifact_bytes,flush_interval_epochs,ingest.run_overrides.allow_reorg_window_settlement, and the ingeststorage.secondary_pathkey, along with theirZINDER_INGEST__*environment variables, before upgrading: ingest refuses to start while any of them is set. Thezinder-ingest verify-canonical-replaysubcommand is removed; it could not open a canonical store written by the release ingest path. The storage benchmark drops its canonical-store range replay scenario and Compose profile, and theZinderStoreReadP95Highalert and its dashboard panel are removed because no release runtime emits the store read metrics they watched. Explorer methods that need canonical storage keep returningFAILED_PRECONDITIONwithDEPENDENCY_NOT_CONFIGURED, as they did before; the Cipherscan compatibility matrix now marks the routes that depend on them. (#115)
Fixed
- Lightwalletd compatibility now refuses to start when its canonical secondary disagrees with the ingest writer fence on the visible block count, which it previously accepted. Native query now fails closed on a writer fence that diverges from its secondary instead of retrying it as replica lag until its attempts run out. Projector, native query, and lightwalletd compatibility now apply the same fence rules. (#116)
- Wallet reads served by
zinder-queryandzinder-compat-lightwalletdnow recordzinder_query_request_*metrics. Before, only operations proxied through the native adapter produced samples, so theZinderWalletQueryP95Highalert and the overview dashboard missed most wallet traffic. Expect new samples for every wallet read, including reads the serving store cannot answer. (#117) - Explorer
/readyznow reportsschema_mismatchorstorage_unavailablewhile its materialized-view catch-up fails, instead of staying ready. Ingest now staysupstream_not_readyuntil its Zebra health probe recovers; the follow loop no longer clears that cause, andWriterStatusreports the probe observation during construction and source recovery. When[node.health]is configured, ingest reportsstartinguntil the first successful probe instead ofready. Ingest, Explorer, and the Cipherscan adapter now reportshutting_downas soon as termination is requested. (#118) - A materialized-view rebuild that resumes without chain-event cursors after a reorg now reverts rows from the replaced chain before rebuilding, instead of keeping them. The transparent address activity index and the address ranking undo journals now write the same bytes regardless of hash-map order or batch size, so replaying the same chain produces identical materialized-view stores. (#119)
- Native
CompactBlocksInRangeand lightwalletdGetBlockRangenow returnNOT_FOUNDwith reasonARTIFACT_UNAVAILABLEfor a range that reaches outside stored history, such as height 0 or a height below the retained floor, instead ofUNAVAILABLE. A missing row inside retained history still reports a storage error, andGetBlockRangeabove the visible tip still returnsOUT_OF_RANGE. (#122) - Subtree-root reads on a checkpointed store now return
NOT_FOUNDwith reasonARTIFACT_UNAVAILABLEfor a root that completed below the retained history and was never loaded, instead ofUNAVAILABLE. A missing root the store loaded, including any root of a store built with the complete subtree-root prefix, still reports a storage error. (#123)
v0.6.0
Immutable
release. Only release title and notes can be modified.
[0.6.0] - 2026-08-10
Changed
- Lightwalletd-compatible deployments now advertise transparent-address support only when their serving data can answer those requests. Startup rejects configurations that lack a transparent wallet projection or retained transaction bytes. (#71)
- Document the production Testnet Zebra readiness calibration that keeps ordinary block gaps available while peer, sync, lag, and stale-tip failures remain fail-closed. (#73)
Fixed
- Stale fixed-view mempool epochs now reach the typed view-expiration path so wallet consumers can reacquire the view without treating expected chain rotation as service unavailability. (#69)
- Secondary query and projection processes no longer create thousands of short-lived storage threads per second while following the chain. (#72)