Post-Join Dependencies

Some annotations and filters require columns from one or more child nodes. A post-join dependency prevents that operation from running until those children have been merged.

Dataframe nodes

Use method_requires with node classes:

from graphreduce.context import method_requires
from graphreduce.node import GraphReduceNode


class OrderNode(GraphReduceNode):
    # Implement the required node methods here.
    ...


class CustomerNode(GraphReduceNode):
    @method_requires(nodes=[OrderNode])
    def do_post_join_annotate(self):
        signup = self.colabbr("signup_date")
        self.df[self.colabbr("days_to_last_order")] = (
            self.df["ord_ts_max"] - self.df[signup]
        ).dt.days

    # Implement the other required node methods here.
    ...

GraphReduce records merged relation classes on the parent. The decorated method returns without executing until every required class is present.

For multiple dependencies:

@method_requires(nodes=[OrderNode, NotificationNode])
def do_post_join_filters(self):
    self.df = self.df[
        self.df["ord_id_count"].notna()
        | self.df["not_id_count"].notna()
    ]

SQL nodes

SQL nodes accept dependency instances as constructor settings:

customers = SQLNode(
    # ...normal node configuration...
    do_post_join_annotate_ops=[...],
    do_post_join_annotate_requires=[orders],
    do_post_join_filters_ops=[...],
    do_post_join_filters_requires=[orders, notifications],
)

The SQL path checks whether columns with each required node's prefix are available before returning the configured operations.

Guidance

• Use pre-join hooks for logic that depends only on the node's own columns.
• Use post-join hooks only when another relation is genuinely required.
• Depend on the smallest set of nodes needed by the expression.
• Reference prefixed columns, because source names have already changed by the time a post-join method runs.
• Cover dependency order with a small graph test; the operation's correctness depends on both graph topology and prefixes.