SharedData

SharedData is a memory-mapped store for records and time series — in the workload category of tick stores — plus the scheduler and worker pool that run the programs producing the data. Two containers do the storing: a table (row-oriented records with a primary key) and a time series (one float matrix per tag, with an index cached by start date). A dependency-aware scheduler and per-host workers do the orchestration; collections, cache and streams sit beside them.

Install the latest published package. The HTTP API is described by the deployed server's OpenAPI document, served at /apidocs.

Why two container types?

SharedData keeps typed data in files that local processes can map through the operating system's page cache. It exists to avoid repeatedly loading, copying and serializing the same working set between cooperating local processes. The data shape determines the container, rather than a particular industry:

Shape and access pattern	Choose	Reason
Keyed records with mixed field types	Table	Update a record by key or append records without imposing a dense time grid
Aligned numeric values by time and named column	Time series	Keep a rectangular float matrix and expose pandas/NumPy views

Shared pages are local to one host. Locking, image refresh, disk persistence and remote access are separate concerns; neither class promises distributed shared memory or a multi-record transaction. Read the synthetic latency results for measured API costs and limits, not a production SLA.

Reading map

Page	What it is
Use cases	The workloads a production deployment runs, generalized
Orchestration	Routines, schedules, the worker pool and the Scheduler's tick
Table	Table, TableDisk, and the SharedNumpy object table() returns
Time series	TimeseriesContainer and TimeSeriesDisk
Logging templates	What the Scheduler logs, and what a routine must not log inside a loop
Swagger	Where the live spec is, and which paths it covers
Quick start	Install, environment, short examples
Configuration	Environment variables

Address

A container is user / database / period / source / name.

• database is a key of DATABASE_PKEYS in src/SharedData/Database.py (MarketData, TimeSeries, Options, …).
• period is D1, M15, M1, or RT.
• source has no /. A table name or a tag has at most one /, and that slash is the partition (PRICES/2026).

from SharedData.SharedData import SharedData

shdata = SharedData(__file__, user='example')
records = shdata.table('MarketData', 'D1', 'DEMO', 'PRICES/2026')
frame = shdata.timeseries('TimeSeries', 'D1', 'DEMO', tag='close', columns=['EXAMPLE'])

table() returns the record array. timeseries(..., tag=...) returns the DataFrame view of that tag. The classes behind those calls are documented on the two core pages.

Client interfaces

Compatible clients can connect through the authenticated HTTP API or MCP. Data access and worker commands remain subject to the installation's permissions.

Project

Install the package, read the changelog, or see Building and publishing. All examples are illustrative.