Read this in 日本語(Japanese)
MarketStore is a database server optimized for financial time-series data. You can think of it as an extensible DataFrame service that is accessible from anywhere in your system, at higher scalability.
It is designed from the ground up to address scalability issues around handling large amounts of financial market data used in algorithmic trading backtesting, charting, and analyzing price history with data spanning many years, and granularity down to tick-level for the all US equities or the exploding crypto currencies space. If you are struggling with managing lots of HDF5 files, this is perfect solution to your problem.
The batteries are included with the basic install - you can start pulling crypto price data from GDAX and writing it to the db with a simple plugin configuration.
MarketStore enables you to query DataFrame content over the network at as low latency as your local HDF5 files from disk, and appending new data to the end is two orders of magnitude faster than DataFrame would be. This is because the storage format is optimized for the type of data and use cases as well as for modern filesystem/hardware characteristics.
MarketStore is production ready! At Alpaca it has been used in production for years in serious business. If you encounter a bug or are interested in getting involved, please see the contribution section for more details.
If you want to get started right away, you can bootstrap a marketstore db instance using our latest docker image. The image comes pre-loaded with the default mkts.yml file and declares the VOLUME /data, as its root directory. To run the container with the defaults:
docker run -i -p 5993:5993 alpacamarkets/marketstore:latestIf you want to run a custom mkts.yml you can create a new container
and load your mkts.yml file into it:
docker create --name mktsdb -p 5993:5993 alpacamarkets/marketstore:latest
docker cp mkts.yml mktsdb:/etc/mkts.yml
docker start -i mktsdbYou can also bind mount
the container to a local host config file: a custom mkts.yml:
docker run -v /full/path/to/mkts.yml:/etc/mkts.yml -i -p 5993:5993 alpacamarkets/marketstore:latestThis allows you to test out the image included plugins with ease if you prefer to skip the copying step suggested above.
By default the container will not persist any written data to your
container's host storage. To accomplish this, bind the data directory to
a local location:
docker run -v "/path/to/store/data:/data" -i -p 5993:5993 alpacamarkets/marketstore:latestOnce data is written to the server you should see a file tree layout like the following that will persist across container runs:
>>> tree /<path_to_data>/marketstore
/<path_to_data>/marketstore
├── category_name
├── WALFile.1590868038674814776.walfile
├── SYMBOL_1
├── SYMBOL_2
├── SYMBOL_3If you have built the cmd package locally, you can open a session with your running docker instance using:
marketstore connect --url localhost:5993MarketStore is implemented in Go, so you can build it from
source pretty easily. You need Go 1.11+ as it uses go mod to manage dependencies.
go get -u github.com/alpacahq/marketstorethen compile and install the project binaries using
make installOptionally, you can install the project's included plugins using
make pluginsYou can also install marketstore using the Homebrew package manager for macOS.
$ brew tap zjhmale/marketstore
$ brew install --HEAD marketstoreTo upgrade marketstore in the future, use upgrade instead of install.
Then you are equipped with marketstore service plist also
$ brew services start marketstore
$ brew services stop marketstoreYou can list available commands by running
marketstore
or
$GOPATH/bin/marketstore
depending on your GOPATH.
You can create a new configuration file named mkts.yml, populated with defaults by running:
$GOPATH/bin/marketstore init
and then start the marketstore server with:
$GOPATH/bin/marketstore start
The output will look something like:
example@alpaca:~/go/bin/src/github.com/alpacahq/marketstore$ marketstore
I0619 16:29:30.102101 7835 log.go:14] Disabling "enable_last_known" feature until it is fixed...
I0619 16:29:30.102980 7835 log.go:14] Initializing MarketStore...
I0619 16:29:30.103092 7835 log.go:14] WAL Setup: initCatalog true, initWALCache true, backgroundSync true, WALBypass false:
I0619 16:29:30.103179 7835 log.go:14] Root Directory: /example/go/bin/src/github.com/alpacahq/marketstore/project/data/mktsdb
I0619 16:29:30.144461 7835 log.go:14] My WALFILE: WALFile.1529450970104303654.walfile
I0619 16:29:30.144486 7835 log.go:14] Found a WALFILE: WALFile.1529450306968096708.walfile, entering replay...
I0619 16:29:30.244778 7835 log.go:14] Beginning WAL Replay
I0619 16:29:30.244861 7835 log.go:14] Partial Read
I0619 16:29:30.244882 7835 log.go:14] Entering replay of TGData
I0619 16:29:30.244903 7835 log.go:14] Replay of WAL file /example/go/bin/src/github.com/alpacahq/marketstore/project/data/mktsdb/WALFile.1529450306968096708.walfile finished
I0619 16:29:30.289401 7835 log.go:14] Finished replay of TGData
I0619 16:29:30.340760 7835 log.go:14] Launching rpc data server...
I0619 16:29:30.340792 7835 log.go:14] Initializing websocket...
I0619 16:29:30.340814 7835 plugins.go:14] InitializeTriggers
I0619 16:29:30.340824 7835 plugins.go:42] InitializeBgWorkers
In order to run MarketStore, a YAML config file is needed. A default file (mkts.yml) can be created using marketstore init. The path to this file is passed in to the start command with the --config flag, or by default it finds a file named mkts.yml in the directory it is running from.
| Var | Type | Description |
|---|---|---|
| root_directory | string | Allows the user to specify the directory in which the MarketStore database resides |
| listen_port | int | Port that MarketStore will serve through for JSON-RPC API |
| grpc_listen_port | int | Port that MarketStore will serve through for GRPC API |
| timezone | string | System timezone by name of TZ database (e.g. America/New_York) |
| log_level | string | Allows the user to specify the log level (info |
| queryable | bool | Allows the user to run MarketStore in polling-only mode, where it will not respond to query |
| stop_grace_period | int | Sets the amount of time MarketStore will wait to shutdown after a SIGINT signal is received |
| wal_rotate_interval | int | Frequency (in mintues) at which the WAL file will be trimmed after being flushed to disk |
| stale_threshold | int | Threshold (in days) by which MarketStore will declare a symbol stale |
| enable_add | bool | Allows new symbols to be added to DB via /write API |
| enable_remove | bool | Allows symbols to be removed from DB via /write API |
| disable_variable_compression | bool | disables the default compression of variable data |
| triggers | slice | List of trigger plugins |
| bgworkers | slice | List of background worker plugins |
root_directory: data
listen_port: 5993
grpc_listen_port: 5995
log_level: info
queryable: true
stop_grace_period: 0
wal_rotate_interval: 5
stale_threshold: 5
enable_add: true
enable_remove: falseAfter starting up a MarketStore instance on your machine, you're all set to be able to read and write tick data.
pymarketstore is the standard python client. Make sure that in another terminal, you have marketstore running
- query data
import pymarketstore as pymkts
param = pymkts.Params('BTC', '1Min', 'OHLCV', limit=10)
cli = pymkts.Client()
reply = cli.query(param)
reply.first().df()shows
Out[5]:
Open High Low Close Volume
Epoch
2018-01-17 17:19:00+00:00 10400.00 10400.25 10315.00 10337.25 7.772154
2018-01-17 17:20:00+00:00 10328.22 10359.00 10328.22 10337.00 14.206040
2018-01-17 17:21:00+00:00 10337.01 10337.01 10180.01 10192.15 7.906481
2018-01-17 17:22:00+00:00 10199.99 10200.00 10129.88 10160.08 28.119562
2018-01-17 17:23:00+00:00 10140.01 10161.00 10115.00 10115.01 11.283704
2018-01-17 17:24:00+00:00 10115.00 10194.99 10102.35 10194.99 10.617131
2018-01-17 17:25:00+00:00 10194.99 10240.00 10194.98 10220.00 8.586766
2018-01-17 17:26:00+00:00 10210.02 10210.02 10101.00 10138.00 6.616969
2018-01-17 17:27:00+00:00 10137.99 10138.00 10108.76 10124.94 9.962978
2018-01-17 17:28:00+00:00 10124.95 10142.39 10124.94 10142.39 2.262249- write data
import numpy as np
import pandas as pd
data = np.array([(pd.Timestamp('2017-01-01 00:00').value / 10**9, 10.0)], dtype=[('Epoch', 'i8'), ('Ask', 'f4')])
cli.write(data, 'TEST/1Min/Tick')
# Out[10]: {'responses': None}
cli.query(pymkts.Params('TEST', '1Min', 'Tick')).first().df()shows
Ask
Epoch
2017-01-01 00:00:00+00:00 10.0- Variable length records
Marketstore is a database that achieves high performance by limiting the number of records in a timeframe to one.
The supported timeframes are from 1D (1 day) to 1Sec (1 second), and basically,
the longer the timeframe is, the faster you can read and write data.
However, it also supports data that does not arrive at a specific interval or more frequently than every second, such as board data and TICK data. Such kind of data is called variable-length records in marketstore.
You can use the variable-length records feature by specifying isvariablelength=True
when you write with pymarketstore.
import numpy as np, pandas as pd, pymarketstore as pymkts
symbol, timeframe, attribute_group = "TEST", "1Sec", "Tick"
data_type = [('Epoch', 'i8'), ('Bid', 'f4'), ('Ask', 'f4'), ('Nanoseconds', 'i4')]
tbk = "{}/{}/{}".format(symbol, timeframe, attribute_group)
client = pymkts.Client()
# --- write variable-length records (=multiple records in a single timeframe (1Sec))
data = np.array([
(pd.Timestamp('2021-01-01 00:00:00').value / 10 ** 9, 10.0, 20.0, 1000000),
(pd.Timestamp('2021-01-01 00:00:00').value / 10 ** 9, 30.0, 40.0, 2000000),
(pd.Timestamp('2021-01-01 00:00:00').value / 10 ** 9, 50.0, 60.0, 3000000),
], dtype=data_type)
client.write(data, tbk, isvariablelength=True)
# --- query variable-length records
params = pymkts.Params(symbol, timeframe, attribute_group)
print(client.query(params=params).first().df())
# --- tearDown
client.destroy(tbk)shows
Bid Ask Nanoseconds
Epoch
2021-01-01 00:00:00+00:00 10.0 20.0 1000000
2021-01-01 00:00:00+00:00 30.0 40.0 2000000
2021-01-01 00:00:00+00:00 50.0 60.0 3000000
Because the data type of Epoch column is always set as 'i8'(=int64) and
it's not sufficient for describing a date with sub-second accuracy,
the sub-second information is stored in another column (=Nanoseconds column, the data type is 'i4') in marketstore.
Connect to a marketstore instance with
// For a local db-
marketstore connect --dir <path>
// For a server-
marketstore connect --url <address>
and run commands through the sql session.
Go plugin architecture works best with Go1.10+ on linux. For more on plugins, see the plugins package Some featured plugins are covered here -
You can receive realtime bars updates through the WebSocket streaming feature. The
db server accepts a WebSocket connection on /ws, and we have built a plugin that
pushes the data. Take a look at the package
for more details.
The batteries are included so you can start pulling crypto price data from GDAX right after you install MarketStore. Then you can query DataFrame content over the network at as low latency as your local HDF5 files from disk, and appending new data to the end is two orders of magnitude faster than DataFrame would be. This is because the storage format is optimized for the type of data and use cases as well as for modern filesystem/hardware characteristics.
You can start pulling data from GDAX if you configure the data poller. For more information, see the package
This plugin allows you to only worry about writing tick/minute level data. This plugin handles time-based aggregation on disk. For more, see the package
You can replicate data from a master marketstore instance to other marketstore instances.
In mkts.yml config file, please set the config as the following:
- master instance
replication:
# when enabled=true, this instance works as master instance and accept connections from replicas
enabled: true
# when tls_enabled=true, transport security between master and replica is enabled.
# tls_enabled: true # both master and replica should have tls_enabled=true to enable TLS
# public/private key pair from a pair of files. The files must contain PEM encoded data.
# The cert file may contain intermediate certificates following the leaf certificate to form a certificate chain.
# cert_file: "/Users/dakimura/projects/misks/tmpcert/server.crt" # both master and replica should have this config to enable TLS
# key_file: "/Users/dakimura/projects/misks/tmpcert/server.key"
# port to be used for the replication protocol
listen_port: 5996
- replica instance(s)
replication:
# when master_host is set, this instance works as a replica instance
master_host: "127.0.0.1:5995"
# when tls_enabled=true on master server, GRPC communication between master and replica is encrypted by SSL.
# tls_enabled: true
# cert_file: "/Users/dakimura/projects/misks/tmpcert/server.crt" # both master and replica should have this config to enable TLS
-
Currently, the replication connection is initialized only at a startup of a marketstore replica instance. Please be sure to start the master instance first when you want to replicate data.
-
Currently, only
writeAPI is supported.deleteAPI result won't be reflected to replica instances. -
When replication is enabled on a replica instance, the instance is set to read-only mode and write API call(s) to the instance will fail.
If you are interested in improving MarketStore, you are more than welcome! Just file issues or requests in github or contact oss@alpaca.markets. Before opening a PR please be sure tests pass-
make unittestWe know the needs and requirements in this space are diverse. MarketStore provides strong core functionality with flexible plug-in architecture. If you want to build your own, look around plugins