ERDDAP defers loading child dataset for EDDGridLonPM180

[SethChampagneNRL]

I am using the EDDGridLonPM180 dataset to correct longitudes in several of our datasets. When I start ERDDAP from a fresh state, my dataset never loads. There is no error, but a log message that ERDDAP is deferring loading the child dataset.

(example snippet is using the sax parser with xinclude)

<?xml version="1.0" encoding="ISO-8859-1"?>
<dataset type="EDDGridLonPM180" datasetID="corrected-dataset-180to180" active="true">
    <dataset type="EDDGridFromNcFiles" datasetID="original-dataset-0to360">
    <!-- dataset definition -->
    </dataset>
</dataset>

Messing with the code, I found the config property forceSynchronousLoading. If I set this to true, then the datasets load correctly. This config property is not accessible to the setup.xml since it looks like it was used for debugging, but I could make the change to add it.

Am I configuring ERDDAP incorrectly? Is this expected behavior? Why does the dataset get deferred on starting ERDDAP?

[rmendels]

@SethChampagneNRL @ChrisJohnNOAA Chris is on leave for a few days. But can I ask a favor because of some other issues I have seen. Can you try the same thing by putting the xml directly in the file and not use xinclude

As for the deferring, despite the message it is not an error. On restart, when ERDDAP™ sees any dataset that it doesn't know about from the cache in bigERDDAP, or a dataset that depends on another dataset loading, it defers dealing with them till all other datasets load and then returns to them. This is so an ERDDAP™ will load as quickly as possible with what it knows, and that new datasets in particular, which can take awhile to load, don't hold up the initial loading.

[SethChampagneNRL]

@rmendels combining the xml file had the same behavior as the xinclude

[rmendels]

@SethChampagneNRL Thanks for testing that. Some users who were having some problems were also using xinclude and I wanted to rule that out as the problem.

[ChrisJohnNOAA]

forceSynchronousLoading was added to allow tests to load datasets more directly. So it's actually quite well tested to have it enabled.

I'm ok with exposing it. However, I do want to also debug why datasets are deferring (this is fine, and expected) and not being loaded after the first pass through the datasets.xml file (this is the problem).

As Roy mentions we defer datasets by default for good reasons, so forceSynchronousLoading should still default to off (and hopefully be rarely needed).

[SethChampagneNRL]

@ChrisJohnNOAA I completely agree. I was unsure if this was a bug or intended behavior, but sounds like a bug

[ChrisJohnNOAA]

I'm working on a change related to issues with the sax parser and child datasets that throw an error (like the deferred loading). It may resolve this issue.

[ChrisJohnNOAA]

@SethChampagneNRL could you try with change #446 and see if it resolves your issue?

If it does fix the loading great, if it doesn't, it's possible we'll need something like the commit here:
f659cb8

to detect when there's an issue with children and defer the parent dataset.

[SethChampagneNRL]

@SethChampagneNRL could you try with change #446 and see if it resolves your issue?

If it does fix the loading great, if it doesn't, it's possible we'll need something like the commit here: f659cb8

to detect when there's an issue with children and defer the parent dataset.

@ChrisJohnNOAA I pulled the ChrisJohnNOAA:side_by_side branch, turned forceSynchronousLoading to false, and wiped my bigParentDirectory to start fresh. I have three datasets configured using EDDGridLonPM180 with child datasets and a fourth configured with plain EDDGridFromNcFiles. I am also using the SAX parser with xinclude and useEddReflection set to true.

On startup, the one EDDGridFromNcFiles dataset loaded immediately, and the three EDDGridLonPM180 were deferred. I had loadDatasetsMinMinutes set to 2 mins, and after 2 mins all four datasets appeared. I tested again with loadDatasetsMinMinutes set to 15 mins, and it took 15 mins for all the datasets to appear. So deferred loading is causing the child datasets to completely skip the initial load cycle.

[ChrisJohnNOAA]

Deferred loading is expected to take some time. There may be an optimization we could do to reload parents after a child dataset loads which might get the dataset loaded sooner. Children datasets don't currently know their parents, so this may be slightly complicated.

That said it sounds like the never load issue is resolved with this change? Are you happy with this being closed with that change, or do you want to leave this open to look into the optimization?

[SethChampagneNRL]

I think "never load" was as red herring since I was waiting 15 mins in my testing for the next dataset load cycle. The real issue is these parent/child datasets get skipped completely on initialization due to the defer logic. If they are nested like in my original snippet example, neither dataset will appear on initialization (unless it was previously initialized in the bigParentDirectory).

[ChrisJohnNOAA]

I'll not mark this as fixed then.

I think the solution to this is what I mentioned above, to detect when all child datasets are loaded and the reload the parent dataset. This can likely be a SAX parser specific implementation (if it needs to be at that level).