Data Structure: Transactions and Holdings

[mkitzmann]

Every user would have a number of transactions which would results in that users holdings. Do we want to save only the transactions and calculate the holdings or do we want to regenerate the holdings each time a transaction is made?
With possibly thousands of transactions it would be very ineffiecent to calculate the holding every time, but with generating them at a fixed time, we do not have a single source of truth, which could eventually lead to an inconsistent state.

[sadortun]

You are right, and it may be a bit more complicated than that.

Few things to consider:

• Holding amounts and $$ and Account values and $$ are only accessible from the user since they are encrypted client-side
• There is a few levels of generalization : Execution -> Order -> Holding -> Account / WatchList -> Global Totals
• Each of theses generalizations requires both to sum up the childs, and allow a direct entry. Example:
  • I have 1000$ in my account right now, and i hold 10 TSLA
  • I bougth 5 more TSLA yesturday at 600$
  • I bougth 3 more TSLA today at 550$
• To update the Account when a Holding is modified we can simply filter/apply when Holding.updatedAt > Account.updatedAt
• When a user delete a Holding, we probably need to only set a flag to false or not allow

I did had a similar thing on a different project and my solution was to handle everything via Time Series

So at the end, you end up with two collections:

• TimeSeries which contains the totals for various durations (year, month, day and maybe hour). FYI Each duration contains data from the smaller time scale. For example, the year Series contains 12 months
• TimeSeriesData which contains the source data for each series. This is useful in the case we need to rebuild the series. We dont have to store everything in one single TimeSeriesData but we need to keep the reference data in an Order or Holding collection for example

After that, its fairly easy to compare the TimeSeriesData.updatedAt > TimeSerie.updatedAt and update the TimeSeries.