Sync patterns guide

Every source table syncs into the Lake through one of seven load patterns. The Lake's built-in guide, Seven ways data lands in your Lake, walks through each pattern with a business example, its trade-offs and an animated demo. Open it while you choose a Sync Type, or any time you want to check what a pattern does with deletes, history and storage.

Open the guide

In the Lake tree, double-click the table, or right-click it and choose Edit.
Open Table > General.
In the Synchronization tile, click Sync patterns guide.

The guide opens over the form. It is a reference dialog: reading it changes nothing on your table. To set a pattern on a table, pick a Sync Type on the same step; see Configure a table and Choose a sync type.

Move around the guide

On a wide screen, the guide's list sits in a 220 px column on the left, under the header Patterns. On a narrow screen, the guide fills the window and the list appears as a row of chips under the title.
The list holds the seven patterns, numbered 1 to 7, and Compare at the end.
Click an item to jump to its section. The list marks the section you are reading as you scroll.
Each card carries its number (Pattern 01 to Pattern 07), the pattern's name and badges, a Use case line, a summary, short notes and an animated example.
Click Close in the corner to leave the guide.

The seven patterns

Each pattern carries two or three badges. Badges that start with LOAD say what each run pulls from the source; badges that start with WRITE say how the rows land in the Lake.

Pattern	Badges	What happens on each run
Full reload	LOAD · ENTIRE TABLE, WRITE · ATOMIC REPLACE	Pulls the entire source table, then swaps it with the Lake table as one atomic operation. If a load fails midway, the old data stays. Deletes propagate: what is gone in the source is gone in the Lake.
Incremental reload	LOAD · LAST X DAYS, WRITE · PARTITION SWAP	Pulls only the lookback window and replaces those partitions as whole blocks, so updates and deletes inside the window are reflected. Outside the window, the pattern is blind: older changes are never picked up.
Full reload · upsert	LOAD · ENTIRE TABLE, WRITE · MERGE BY KEY, DELETES NEVER PROPAGATE	Pulls the entire table and merges by key: existing rows update, new rows insert. Rows deleted in the source stay in the Lake; once a record lands, it stays.
Incremental · upsert	LOAD · LAST X DAYS, WRITE · MERGE BY KEY, DELETES NEVER PROPAGATE	The window pull of incremental reload, merged by key instead of swapped. A row deleted in the source inside the window stays in the Lake, where incremental reload would drop it.
Full snapshot	LOAD · ENTIRE TABLE, WRITE · APPEND VERSION, SNAPSHOT FROM / TO	Appends a full copy of the table as a new version on every run, so you can read it as it looked on a past date. Storage grows by the full table's size on every run.
Incremental snapshot	LOAD · LAST X DAYS, WRITE · APPEND VERSION, SNAPSHOT FROM / TO	Versions only the part of the table the window covers, so a large table keeps change history without copying its untouched rows.
Onetime	LOAD · ONLY WHEN EMPTY, WRITE · INSERT ONCE	Loads the table only while it is empty, and skips every later run. To load it again, truncate the Lake table (see Run, stop and troubleshoot loads).

A few things the cards add:

Examples. Each card pairs the pattern with a business case: a customers dimension that must mirror the CRM, a transactions table where only recent days change, orders that must survive source purges, and a status lookup defined once.
Notes. Short notes flag the trade-offs, such as the all-or-nothing swap of a full reload and the storage cost of a full snapshot.
Empty tables. The Full reload on empty note on the incremental reload card says an empty Lake table triggers one automatic full load before incremental takes over. That first load is controlled by Full Reload When Empty on Table > General; see Choose a sync type for how it is sized.

Note

Snapshot patterns version rows with Snapshot From and Snapshot To. The current version shows 2099-12-31 23:59:59.999 in Snapshot To, and a previous version is closed at exactly one millisecond before the next snapshot starts, so version ranges never overlap and never leave a gap.

The cheat sheet

The last item in the guide's list, Compare, holds the cheat sheet Which pattern do I need?. It compares all seven patterns in one table and colours the answers that flag a caveat.

Pattern	Pulled per run	Deletes propagate	Keeps history	Storage growth	Needs date column
1 · Full reload	Entire table	Yes	No	Flat	No
2 · Incremental reload	Last X days	Only in window	No	Flat	Yes
3 · Full upsert	Entire table	Never	Deleted rows only	Slow	No
4 · Incremental upsert	Last X days	Never	Deleted rows only	Slow	Yes
5 · Full snapshot	Entire table	Never	Every version	Fast - size × runs	No
6 · Incremental snapshot	Last X days	Never	Recent versions	Moderate	Yes
7 · Onetime	Once, then nothing	-	No	Flat	No

Patterns and sync types

The guide uses short names; the Sync Type list on a table uses the full ones. They match like this:

Guide card	Sync Type on a table
Full reload	Full Reload (Truncate Insert)
Incremental reload	Incremental Reload (Truncate Insert)
Full reload · upsert	Full Reload (Upsert)
Incremental · upsert	Incremental (Upsert)
Full snapshot	Full Snapshot
Incremental snapshot	Incremental Snapshot
Onetime	Onetime

The loop type, Incremental Reload Loop (Truncate Insert), has no card of its own in the guide: it is the loop variant of incremental reload and walks the window in steps of Loop Increment. Set a type on Configure a table, and compare them all on Choose a sync type.

Questions, answered

Where do I find the sync patterns guide?

Open any table's popup and go to Table > General. In the Synchronization tile, click Sync patterns guide. The button sits with the sync settings, so you can open the guide while you pick a Sync Type.

What do the badges on each pattern mean?

Badges that start with LOAD say what each run pulls from the source: the entire table, the last X days, or only while the table is empty. Badges that start with WRITE say how rows land in the Lake: atomic replace, partition swap, merge by key, append version or insert once. Two extra badges add detail: DELETES NEVER PROPAGATE on the upsert patterns and SNAPSHOT FROM / TO on the snapshot patterns.

How do I compare all seven patterns at once?

Open Compare, the last item in the guide's list. It shows the Which pattern do I need? cheat sheet: every pattern on one row, with what it pulls per run, whether deletes propagate, whether it keeps history, how storage grows and whether it needs a date column.

How do I move around the guide?

Click a pattern in the list to jump to it; on a narrow screen, click a chip at the top instead. The list highlights the section you are reading as you scroll.

Are the examples in the guide real data?

No. The animations use example tables and fictional sample rows to show what each pattern does over one or two runs; they do not show your data.

How do the guide's patterns relate to the sync types on a table?

They are the same seven families under shorter names: Full reload is Full Reload (Truncate Insert), Full reload · upsert is Full Reload (Upsert), and so on. The table in Patterns and sync types lists every pair.

What does the DELETES NEVER PROPAGATE badge mean?

On the two upsert patterns, a row deleted in the source is never deleted in the Lake. The merge is by key, so once a record has landed it stays, even if the source later deletes it. The card's note flags this as the feature to check first: confirm that it is what you want before a table uses an upsert.

Why does the incremental reload example keep a row the source deleted?

Because incremental reload reads only the lookback window. A delete that happened outside the window is never seen, so the row stays in the Lake; the example's caption calls this out. Choose a sync type explains how to widen the window or switch to a full pattern.

Why do snapshot tables show a 2099 date?

Snapshot patterns version rows with Snapshot From and Snapshot To. The current version keeps 2099-12-31 23:59:59.999 in Snapshot To until a newer snapshot arrives; the previous version is then closed one millisecond before the new one begins. That keeps version ranges clean, with no overlaps and no gaps. To read a past version, filter on the two columns; Choose a sync type shows how.

Is the loop sync type in the guide?

The guide has seven cards and none of them is the loop. Incremental Reload Loop (Truncate Insert) is the loop variant of incremental reload: it walks the same window in steps of Loop Increment. Choose a sync type covers it.