Does sortby filter search results?

[bmcandr]

When searching Planetary Computer's STAC I noticed that the sortby query parameter appears to filter search results. Is this expected? If so, what is the cause?

For example, the following search returns 150 items:

import planetary_computer
import pystac_client

client = pystac_client.Client.open(
    "https://planetarycomputer.microsoft.com/api/stac/v1",
    modifier=planetary_computer.sign_inplace
)

search = client.search(
    collections=["sentinel-2-l2a"],
    datetime="2021-12-17/2022-12-17",
    query={
        "s2:mgrs_tile": {"eq": "50QMJ"}
    }
)

len(search.item_collection())
>>> 150

while the same search with sortby on the eo:cloud_cover property returns only 148 Items:

search = client.search(
    collections=["sentinel-2-l2a"],
    datetime="2021-12-17/2022-12-17",
    query={
        "s2:mgrs_tile": {"eq": "50QMJ"}
    },
    sortby="eo:cloud_cover",
)

len(search.item_collection())
>>> 148

[gadomski]

🤔 I can't reproduce, using a modified version of your script intended to discover whether the missing ids are stable:

import planetary_computer
import pystac_client

client = pystac_client.Client.open(
    "https://planetarycomputer.microsoft.com/api/stac/v1",
    modifier=planetary_computer.sign_inplace,
)

search = client.search(
    collections=["sentinel-2-l2a"],
    datetime="2021-12-17/2022-12-17",
    query={"s2:mgrs_tile": {"eq": "50QMJ"}},
)
ids_without_sort = set()
for item in search.items():
    ids_without_sort.add(item.id)
print(f"Without sort, {len(ids_without_sort)} items")

for _ in range(5):
    search = client.search(
        collections=["sentinel-2-l2a"],
        datetime="2021-12-17/2022-12-17",
        query={"s2:mgrs_tile": {"eq": "50QMJ"}},
        sortby="eo:cloud_cover",
    )
    ids_with_sort = set()
    for item in search.items():
        ids_with_sort.add(item.id)
    missing_ids = sorted(ids_without_sort - ids_with_sort)
    if missing_ids:
        print(f"With sort, {len(ids_with_sort)} items: {', '.join(missing_ids)}")
    else:
        print(f"With sort, {len(ids_with_sort)} items, none missing")

Output:

Without sort, 150 items
With sort, 150 items, none missing
With sort, 150 items, none missing
With sort, 150 items, none missing
With sort, 150 items, none missing
With sort, 150 items, none missing

[bmcandr]

Hm. I was able to reproduce my earlier result on PCH with a fresh notebook and using the base image/env.

And confirmed with your modified version:

The "missing" Items appear to be stable:

• S2A_MSIL2A_20220503T023551_R089_T50QMJ_20220503T164728
• S2A_MSIL2A_20221102T024851_R132_T50QMJ_20221102T174808

[gadomski]

Very strange that we're getting different results. One question and one ask:

• What's the eo:cloud_cover attribute of those missing items?
• Can you update pystac-client to the latest version, v0.7.5? Unlikely that it's the problem but have to check.

[bmcandr]

Cloud cover values for the missing items:

S2A_MSIL2A_20221102T024851_R132_T50QMJ_20221102T174808 -- 93.301648
S2A_MSIL2A_20220503T023551_R089_T50QMJ_20220503T164728 -- 93.692213

---

After updating to v0.7.5 I cannot reproduce my earlier result using the code above. Looks like the most recent change to ItemSearch behavior removed the default limit sent to the server. Passing limit=100 to recreate the earlier behavior of pystac_client does reproduce it:

[gadomski]

Yup, I can reproduce:

import planetary_computer
import pystac_client

print(f"pystac-client version: {pystac_client.__version__}")


client = pystac_client.Client.open(
    "https://planetarycomputer.microsoft.com/api/stac/v1",
    modifier=planetary_computer.sign_inplace,
)


def get_ids(use_sort: bool, use_limit: bool) -> set[str]:
    if use_sort:
        sortby = "eo:cloud_cover"
    else:
        sortby = None
    if use_limit:
        limit = 100
    else:
        limit = None
    search = client.search(
        collections=["sentinel-2-l2a"],
        datetime="2021-12-17/2022-12-17",
        query={"s2:mgrs_tile": {"eq": "50QMJ"}},
        sortby=sortby,
        limit=limit,
    )
    ids = set()
    for item in search.items():
        ids.add(item.id)
    return ids


ids = get_ids(use_sort=False, use_limit=False)
print(f"Without sort, without limit, {len(ids)} items")


def assess(use_sort: bool, use_limit: bool) -> None:
    new_ids = get_ids(use_sort=use_sort, use_limit=use_limit)
    missing_ids = sorted(ids - new_ids)
    label = (
        f"With{'' if use_sort else 'out'} sort, with{'' if use_limit else 'out'} limit"
    )
    if missing_ids:
        print(f"{label}, {len(new_ids)} items: {', '.join(missing_ids)}")
    else:
        print(f"{label}, {len(new_ids)} items, none missing")


assess(use_sort=True, use_limit=False)
assess(use_sort=False, use_limit=True)
assess(use_sort=True, use_limit=True)

Output:

pystac-client version: 0.7.5
Without sort, without limit, 150 items
With sort, without limit, 150 items, none missing
Without sort, with limit, 150 items, none missing
With sort, with limit, 148 items: S2A_MSIL2A_20220503T023551_R089_T50QMJ_20220503T164728, S2A_MSIL2A_20221102T024851_R132_T50QMJ_20221102T174808

I've seen issues like this before with pgstac, I'll open an issue over there and see what they say. Thanks for the report!

[gadomski]

There's a similar/same issue reported on the Planetary Computer, who will likely be the folks to drive the fix in pgstac: microsoft/PlanetaryComputer#301. I've linked this discussion in over there, I'd watch that issue for updates.