Choosing a time series database

[goodboy]

Did some decent starter research on what systems will work for production grade data feed collaboration.

---

I'm going to list some top picks and what I deem to be pros/cons; comes with a check-mark if we have an existing integration.

• [x] alpaca's marketstore
  • built for use with numpy
  • it seems simple, with a reasonably manageable code base in golang; hackable and with an interesting design
  • recently got grpc support
  • plugin system which seems to a a few decent things you might not find in your average TSDB:
  • a websocket streaming interace on updates
  • an on disk aggregator
  • no clustering, failover, sharding

---

on the suit fronts

• [ ] #495

  • general roadmap dash: https://roadmap.databento.com/roadmap
  • docs: https://docs.databento.com/
  • symbology API (#467!): https://docs.databento.com/api-reference-historical/symbology
    • for futes: https://roadmap.databento.com/b/n0o5prm6/feature-ideas/smart-symbology-lead-month-contracts
    • WIP for opts: https://roadmap.databento.com/b/n0o5prm6/feature-ideas/smart-symbology-for-options
    • treasuries data: https://roadmap.databento.com/b/n0o5prm6/feature-ideas/cash-treasuries-data
  • [ ] live API clients and avail src codez: https://github.com/databento/
    • python client: https://github.com/databento/databento-python
      • docs: https://docs.databento.com/getting-started?historical=python&live=python
      • api docs: https://docs.databento.com/api-reference-live
    • rust backend lib: https://github.com/databento/dbn
      • already opened an issue asking about running you're own: https://github.com/databento/databento-python/issues/9

• [ ] the new arcticdb, a replacement for the prior arctic project (see below)

  • repo: https://github.com/man-group/ArcticDB
  • Feb 2023 update about the release of the C++ rewrite in collab with bloomy: https://www.man.com/man-group-brings-powerful-dataframe-database-product-arcticdb-to-market-with-bloomberg
  • [ ] need to figure out if we can store np.ndarrays as well or potentially write a polars adapter?

• [ ] the original arctic by the "man group" which allows serializing dfs and np arrays.

  • seems to have competed more directly with marketstore but is built on mongodb
  • has a tick storage format using pd.DataFrame: https://arctic.readthedocs.io/en/latest/tickstore/
  • talk on storing 10^12 rows of timeseries: https://vimeo.com/showcase/3660528/video/145842301
  • cryptofeed backend impl: https://github.com/bmoscon/cryptofeed/blob/6cf185de641cd63b180f2a34cf52c0773e5961a3/cryptofeed/backends/arctic.py

• [ ] QuestDb which seems to claim QuestDB is the fastest open source time series database

  • but it's java...
  • repo: https://github.com/questdb/questdb
  • cryptofeed backend impl: https://github.com/bmoscon/cryptofeed/blob/6cf185de641cd63b180f2a34cf52c0773e5961a3/cryptofeed/backends/quest.py

• [ ] greptimedb: https://github.com/GreptimeTeam/greptimedb

  An open-source, cloud-native, distributed time-series database with PromQL/SQL/Python supported.

  • written in rust

---

legacy-ish FOSS projects that probably aren't specialized enough..

• influxdb is the gold standard in TSDBs

  • here's a quant related SO question that resulted in this pump
  • the github repo
  • cryptofeed backend impl: https://github.com/bmoscon/cryptofeed/blob/6cf185de641cd63b180f2a34cf52c0773e5961a3/cryptofeed/backends/influxdb.py

• timescaledb is an extension on top of postgres

  • here's a 18' pump on pushing 3B points a day and how it outperforms influxdb
  • another pump blog post on ingesting from kafka
  • has all the benefits of postgres
  • afaik can work with apache arrow and turbodbc

---

Additional Resources

• blog post on connecting apache arrow to the SQL world
• a post from uber on moving from postgres to mysql
• cryptofeed's long list of backends which includes some serious projects i hadn't seen before.

---

Thots

• One of the things I want to keep in mind is supporting apache arrow formatted data for streaming. The arrow project's new flight system seems to be the way things are going.

[goodboy]

Probably worth reading up on some professional grade systems:

• kdb+
  • which is a column oriented db
  • there's a free version which requires being on-line

[goodboy]

Did a little more research on timescaledb and found this reddit thread which criticized its performance :crying_cat_face:

On the brighter side, Ameobea suggested the very cool looking tectonicdb for L2 tick data:

• example python client
• the socket spec
• L2 book plotting code discussed in detail in this blog post.
• oh, and it's got an offline parser which can be used to read dataframes, yeye!

[goodboy]

Noticing a category of "IoT" capture dbs:

• griddb our of japan

[goodboy]

https://github.com/singularity-data/risingwave