Skip to content

Move the Overture pin off the release about to be pruned - #125

Open
bkeepers wants to merge 1 commit into
mainfrom
overture-2026-07-22
Open

Move the Overture pin off the release about to be pruned#125
bkeepers wants to merge 1 commit into
mainfrom
overture-2026-07-22

Conversation

@bkeepers

Copy link
Copy Markdown
Member

Overture publishes to its public bucket on a roughly monthly cadence and keeps only the last two releases, deleting the rest. The pin sat on the older of the two, so the day the next release lands, 2026-06-17.0 disappears and every inland-water-mask build fails in _check_release with nothing to fall back to. Moving to the newer release buys a full cycle of headroom.

This is a pin refresh, not a data change. The schema the window read depends on is unchanged: subtype, class, is_salt, is_intermittent and the source_tags $.tidal passthrough all still exist and are still populated (Thames estuary window: 109 tidal=yes, 37 tidal=no, so the undocumented passthrough survives another release). The partition is the same shape, 32 parquet parts and 26.2 GiB, and the two releases return identical ids over the probed windows.

watermask's version= moves with it. The profile's rerun-triggers deliberately exclude code, so editing landmask.py invalidates nothing on its own; without the param bump the box would keep serving the mask built from the old release and the pin would only take effect the next time something else forced a rebuild.

Rollout

Merging this makes the next sources run rebuild the mask from 2026-07-22.0. That run stages the new partition (26 GB, into store/landmask/overture-2026-07-22.0/ beside the existing cache rather than over it) and rebuilds water.fgb and land-z8.tif, so it is a longer sources run than usual. The mask's own gates decide whether it publishes: the conservation count and the 494 required lakes both have to pass, which is what makes a release bump safe to take without hand-inspecting the result.

Worth doing deliberately rather than letting it ride: the current mask is the one the v2026.08.10 tiles were built against, and water-pre-huron-fix.fgb on R2 is the rollback if the new release's water differs more than expected.

Overture keeps the last two releases in the public bucket and deletes the rest,
and the pin sat on the older of the two. The next release lands on a roughly
monthly cadence (the two published are 35 days apart), and the day it does,
2026-06-17.0 disappears and every inland-water-mask build dies in
_check_release with nothing to fall back to. Pinning the newer one buys a full
release cycle of headroom without changing what the mask is built from.

The schema the window read depends on is unchanged: subtype, class, is_salt,
is_intermittent and the source_tags '$.tidal' passthrough all still exist and
are still populated (Thames estuary window: 109 tidal=yes, 37 tidal=no; the
undocumented passthrough survives another release). The partition is the same
shape as before, 32 parquet parts and 26.2 GiB, and the two releases return
identical ids over the probed windows, so this is a pin refresh rather than a
data change.

watermask's version= goes with it. The profile's rerun-triggers deliberately
exclude code, so an edit to landmask.py invalidates nothing on its own: without
the param bump the box would keep serving the mask built from the old release
and the pin would only take effect the next time something else forced a
rebuild.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant