Skip to content

Latest commit

 

History

History
418 lines (226 loc) · 19.8 KB

CHANGELOG.md

File metadata and controls

418 lines (226 loc) · 19.8 KB
Raw

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

v4.0.0-rc.1

Fixes

  • Fix an issue preventing the setup command from running on a clickhouse backend because of the reorg settings.

v4.0.0-beta

Highlights

  • This release brings support for managing reorgs in Postgres database, enabled by default when --undo-buffer-size to 0.

Breaking changes

  • A change in your SQL schema may be required to keep existing substreams:SQL integrations working:

    • The presence of a primary key (single key or composite) is now MANDATORY on every table.
    • The sf.substreams.sink.database.v1.TableChange message, generated inside substreams, must now exactly match its primary key with the one in the SQL schema.
    • You will need to re-run setup on your existing PostgreSQL databases to add the substreams_history table. You can use the new --system-tables-only flag to perform only that.

  • Since reorgs management is not yet supported on Clickhouse, users will have to set --undo-buffer-size to a non-zero value (12 was the previous default)

Protodefs v1.0.4

v3.0.5

  • Fixed regression: run command was incorrectly only processing blocks staying behind the "FinalBlocks" cliff.

v3.0.4

  • Fixed support for tables with primary keys misaligned with database_changes's keys (fixing Clickhouse use case)

Protodefs v1.0.3

  • Added support for selecting engine postgres or clickhouse to sinkconfig protobuf definition

v3.0.3

  • Fixed missing uint8 and uint16 cast for clickhouse driver conversion
  • Bump version of schema dependency to fix errors with Clickhouse 23.9 enums being updated to String representation and not numbers.

v3.0.2

  • Fixed default endpoint sfor networks 'goerli' and 'mumbai'
  • Added --postgraphile flag to setup, which will add a @skip comment on cursor table so Postgraphile doesn't try to serve cursors (it resulted in a name collision with Postgraphile internal names)
  • Fixed a bug with Clickhouse driver where different integer sizes need explicit conversion

v3.0.1

Fixed

  • Fixed an issue where the schema encoded in the SinkConfig part of a manifest would not be encoded correctly, leading to garbled (base64) bytes being sent to the SQL server instead of the schema.

v3.0.0

Highlights

This release brings a major refactoring enabling support for multiple database drivers and not just Postgres anymore. Our first newly supported driver is Clickhouse which defines itself as The fastest and most resource efficient open-source database for real-time apps and analytics. In the future, further database driver could be supported like MySQL, MSSQL and any other that can talk the SQL protocol.

Now that we support multiple driver, keeping the substreams-sink-postgres didn't make sense anymore. As such, we have renamed the project from substreams-sink-postgresql to substreams-sink-sql since it now supports Clickhouse out of the box. The binary and Go modules have been renamed in consequence.

Another major change brought by this release is the usage of Substreams "Deployable Unit". What we call a "Deployable Unit" is a Substreams manifest that fully defines a deployment packaged as a single artifact. This change how the sink is operated; the SQL schema, output module and "Network" identifer are now passed in the "SinkConfig" section of the Substreams manifest instead of being accepted at command line.

Read the Operators section below to learn how to migrate to this new version.

Operators

Passing the schema and the module_name to the run and setup commands is no longer accepted via arguments, they need to be written to the substreams.yaml file.

Before:

substreams-sink-sql setup "psql://..." "path/to/schema.sql"
substreams-sink-sql run "psql://..." mainnet.eth.streamingfast.io:443 https://github.com/streamingfast/substreams-eth-block-meta/releases/download/v0.5.1/substreams-eth-block-meta-v0.5.1.spkg db_out [<range>]

Now:

  • Create a deployable unit file, let's call it substreams.prod.yaml with content:
specVersion: v0.1.0
package:
  name: "<name>"
  version: v0.0.1

imports:
  sql: https://github.com/streamingfast/substreams-sink-sql/releases/download/protodefs-v1.0.1/substreams-sink-sql-protodefs-v1.0.1.spkg
  main: https://github.com/streamingfast/substreams-eth-block-meta/releases/download/v0.5.1/substreams-eth-block-meta-v0.5.1.spkg

network: mainnet

sink:
  module: main:db_out
  type: sf.substreams.sink.sql.v1.Service
  config:
    schema: "./path/to/schema.sql"

In this <name> is the same name as what <manifest> defines was, https://github.com/streamingfast/substreams-eth-block-meta/releases/download/v0.5.1/substreams-eth-block-meta-v0.5.1.spkg is the current manifest you deploy.

The ./path/to/schema.sql would point to your schema file (path resolved relative to parent directory of substreams.prod.yaml).

The 'network: mainnet' will be used to resolve to an endpoint. You can configure each network to have its own endpoint via environment variables SUBSTREAMS_ENDPOINTS_CONFIG_<NETWORK> or override this mechanism completely by using the --endpoint flag. Most used networks have default endpoints.

  • Setup your database:
substreams-sink-sql setup <dsn> substreams.prod.yaml
  • Run the sink:
substreams-sink-sql run <dsn> substreams.prod.yaml

Similar changes have been applied to other commands as well.

v2.5.4

Added

  • Added average flush duration to sink stats.

  • Added log line when flush time to database is > 5s in INFO and in WARN if > 30s.

Fixed

  • Fixed pprof HTTP routes not properly registered.

Changed

  • Renamed metric Prometheus metric substreams_sink_postgres_flushed_entries_count to substreams_sink_postgres_flushed_rows_count, adjust your dashboard if needed and change it found a Gauge to a Counter.

v2.5.3

  • Refactored internal code to support multiple database drivers.

  • Experimental clickhouse is now supported as a new clickhouse is now supported* Added driver abstraction

    You can connect to Clickhouse by using the following DSN:

    • Not encrypted: clickhouse://<host>:9000/<database>?username=<user>&password=<password>
    • Encrypted: clickhouse://<host>:9440/<database>?secure=true&skip_verify=true&username=<user>&password=<password>

    If you want to send custom args to the connection, you can use by sending as query params.

v2.5.2

Changed

  • Bumped logging library to latest version which should fixed problem where containerized workload are not printing logs out in JSON format.

v2.5.1

This is a bug fix release containing a fix for inserting rows into a table for which no primary key constraint exist. For now, we still requires internally that your provide an id in your DatabaseChange of your row, a future update will lift that limitations.

v2.5.0

Highlights

This releases brings improvements to reported progress message while your Substreams executes which should greatly enhanced progression tracking

Note

Stay tuned, we are planning even more useful progression tracking now that we've updated progression data sent back to the client!

This releases also introduces a new mode to dump data in the database at high speed, useful for large amount of data insertion.

Substreams Progress Messages

Bumped substreams-sink v0.3.1 and substreams to v1.1.12 to support the new progress message format. Progression now relates to stages instead of modules. You can get stage information using the substreams info command starting from version v1.1.12.

Important

This client only support progress messages sent from a server using substreams version >=v1.1.12

Changed Prometheus Metrics

  • substreams_sink_progress_message removed in favor of substreams_sink_progress_message_total_processed_blocks
  • substreams_sink_progress_message_last_end_block removed in favor of substreams_sink_progress_message_last_block (per stage)

Added Prometheus Metrics

  • Added substreams_sink_progress_message_last_contiguous_block (per stage)
  • Added substreams_sink_progress_message_running_jobs(per stage)

New injection method

A new injection method has been added to this substreams-sink-postgres release. It's a 2 steps method that leverage COPY FROM SQL operations to inject at high speed a great quantity of data.

Note

This method will be useful if you insert a lot of data into the database. If the standard ingestion speed satisfy your needs, continue to use it, the new feature is an advanced use case.

See the High Throughput Injection section of the README.md file to check how to use it.

Added

  • Added newer method of populating the database via CSV (thanks @gusinacio!).

    Newer commands:

    • generate-csv: Generates CSVs for each table
    • inject-csv: Injects generated CSV rows for <table>

v2.4.0

Changed

  • gRPC InvalidArgument error(s) are not retried anymore like specifying and invalid start block or argument in your request.

  • Breaking Flag shorthand -p for --plaintext has been re-assigned to Substreams params definition, to align with substreams run/gui on that aspect. There is no shorthand anymore for --plaintext.

    If you were using before -p, please convert to --plaintext.

    Note We expect that this is affecting very few users as --plaintext is usually used only on developers machine.

Added

  • Added support for --params, -p (can be repeated multiple times) on the form -p <module>=<value>.

v2.3.4

Added

  • Added logging of new Session received values (linear_handoff_block, max_parallel_workers and resolved_start_block).

  • Added --header, -H (can be repeated multiple times) flag to pass extra headers to the server.

Changed

  • Now reporting available columns when an unknown column is encountered.

v2.3.3

Fixed

  • Batches written to the database now respects the insertion ordering has received from your Substreams. This fixes for example auto-increment to be as defined on the chain.

v2.3.2

Fixed

  • Fixed problem where string had unicode character and caused pq: invalid message format.

v2.3.1

Fixed

  • The substreams-sink-postgres setup command has been fixed to use the correct schema defined by the DSN.

  • The cursors table suggestion when the table is not found has been updated to be in-sync with table used in substreams-sink-postgres setup.

Changed

v2.3.0

Added

  • Added Composite keys support following the update in substreams-database-change

    The code was updated to use oneOf primary keys (pk and composite) to keep backward compatibility. Therefore, Substreams using older versions of DatabaseChange can still use newer versions of postgres-sink without problems. To use composite key, define your schema to use Postgres composite keys, update to latest version of substreams-database-changes and update your code to send a CompositePrimaryKey key object for the primary_key field of the TableChange message.

  • Added escape to value in case the postgres data type is BYTES. We now escape the byte array.

Fixed

  • Added back support for old Substreams Database Change Protobuf package id sf.substreams.database.v1.DatabaseChanges.

v2.2.1

Changed

  • Reduced the amount of allocations and escaping performed which should increase ingestion speed, this will be more visible for Substreams where a lot of entities and columns are processed.

Fixed

  • The schema is correctly respected now for the the cursors table.

v2.2.0

Highlights

Cursor Bug Fix

It appeared that the cursor was not saved properly until the first graceful shutdown of substreams-sink-postgres. Furthermore, the on exit save was actually wrong because it was saving the cursor without flushing accumulated data which is wrong (e.g. that we had N blocks in memory unflushed and a cursor, and we were saving this cursor to the database without having flushed the in memory logic).

This bug has been introduced in v2.0.0 by mistake which means if we synced a new database with v2.0.0+, there is a good chance your are actually missing some data in your database. It's highly recommended that you re-synchronize your database from scratch.

Note If your are using the same .spkg that you are using right now, database ingestion from scratch should go at very high speed because you will be reading from previously cached output, so the bottleneck should be network and the database write performance.

Behavior on .spkg update

In the release, we change a big how cursor is associated to the <module>'s hash in the database and how it's stored.

Prior this version, when loading the cursor back from the database on restart, we were retrieving the cursor associated to the <module>'s hash received by substreams-sink-postgres run. The consequence of that is that if you change the .spkg version you were sinking with, on restart we would find no cursor since the module's hash of this new .spkg would have changed and which you mean a full sync back would be happening because we would start without a cursor.

This silent behavior is problematic because it could seen like the cursor was lost somehow while actually, we just picked up a new one from scratch because the .spkg changed.

This release brings in a new flag substreams-sink-postgres run --on-module-hash-mistmatch=error (default value shown) where it would control how we should react to a changes in the module's hash since last run.

  • If error is used (default), it will exit with an error explaining the problem and how to fix it.
  • If warn is used, it does the same as 'ignore' but it will log a warning message when it happens.
  • If ignore is set, we pick the cursor at the highest block number and use it as the starting point. Subsequent updates to the cursor will overwrite the module hash in the database.

There is a possibility that multiple cursors exists in your database, hence why we pick the one with the highest block. If it's the case, you will be warned that multiple cursors exists. You can run substreams-sink-postgres tools cursor cleanup <manifest> <module> --dsn=<dsn> which will delete now useless cursors.

The ignore value can be used to change to a new .spkg while retaining the previous data in the database, the database schema will start to be different after a certain point where the new .spkg became active.

Added

  • Added substreams-sink-postgres run --on-module-hash-mistmatch=error to control how a change in module's hash should be handled.

Changed

  • Changed behavior of how cursor are retrieved on restart.

Fixed

  • Fixed cursor not being saved correctly until the binary exits.

  • Fixed wrong handling of updating the cursor, we were not checking if a row was updated when doing the flush operation.

  • Fixed a bug where it was possible if the sink was terminating to write a cursor for data not yet flushed. This was happening if the substreams-sink-postgres run was stopped before we ever written a cursor, which normally happens each 1000 blocks. We don't expect anybody to have been hit by this but if you are unsure, you should check data for the 1000 first blocks of you sink (for example from 11 000 000 to 11 001 000 if your module start block was 11 000 000).

v2.1.0

Changed

  • Column's schema type that are not known by the sql library we know will now be transferred as-is to the database.

    There is a lot of column's for which the sql library we use have to Go representation for by default. This is the case for example for the numeric column's type. Previously, this would be reported directly as an error, Now, we pass the received value from your Substreams unmodified to the database engine. It will be your responsibility to send the data in the right format accepted by the database. We send the value as-is, without escaping and without sanitization, so this is a risk if you don't control the Substreams.

Added

  • When doing substreams-sink-postgres run, the <manifest> argument now accepts directory like ..

Fixed

  • Fixed timestamp received in RFC3339 format.

v2.0.2

Changed

  • Diminish amount of allocations done to perform fields transformation.

Fixed

  • Fixed some places where escaping for either identifier or value was not done properly.

  • Fixed double escaping of boolean values.

v2.0.1

Added

  • Added proper escaping for table & column names to allow keyword column names to use keywords as column names such as to and from etc.

v2.0.0

Highlights

This release drops support for Substreams RPC protocol sf.substreams.v1 and switch to Substreams RPC protocol sf.substreams.rpc.v2. As a end user, right now the transition is seamless. All StreamingFast endpoints have been updated to to support the legacy Substreams RPC protocol sf.substreams.v1 as well as the newer Substreams RPC protocol sf.substreams.rpc.v2.

Support for legacy Substreams RPC protocol sf.substreams.v1 is expected to end by June 6 2023. What this means is that you will need to update to at least this release if you are running substreams-sink-postgres in production. Otherwise, after this date, your current binary will stop working and will return errors that sf.substreams.v1.Blocks is not supported on the endpoint.

From a database and operator standpoint, this binary is fully backward compatible with your current schema. Updating to this binary will continue to sink just like if you used a prior release.

Retryable Errors

The errors coming from Postgres are not retried anymore and will stop the binary immediately.

Operators

If you were using environment variable to configure the binary, note that the environment prefix has changed from SINK_ to SINK_POSTGRES_.

Changed

  • Deprecated The flag --irreversible-only is deprecated, use --final-blocks-only instead.

Added

  • Added command substreams-sink-postgres tools --dsn <dsn> cursor read to read the current cursors stored in your database.

  • Dangerous Added command substreams-sink-postgres tools --dsn <dsn> cursor write <module_hash> <cursor> to update the cursor in your database for the given <module_hash>

    Warning This is a destructive operation, be sure you understand the consequences of updating the cursor.

  • Dangerous Added command substreams-sink-postgres tools --dsn <dsn> cursor delete [<module_hash>|--all] to delete the cursor associated with the given module's hash or all cursors if --all is used.

    Warning This is a destructive operation, be sure you understand the consequences of updating the cursor.

v1.0.0

Highlights

This is the latest release before upgrading to Substreams RPC v2.

Added

  • Added --infinite-retry to never exit on error and retry indefinitely instead.

  • Added --development-mode to run in development mode.

    Warning You should use that flag for testing purposes, development mode drastically reduce performance you get from the server.

  • Added --irreversible-only to only deal with final (irreversible) blocks.