Error of INDEL mode during VQSR process

Hello,
I'm trying to do VQSR on exome data of 50,000 samples. Since this dataset is too big, I used GenomicsDBImport to merge. Whole exome merging is also slower than expected, so I did this per chromosome.
SNP mode for VQSR went on smoothly while the INDEL mode has several problems.
I used the command below at first:

```
time $gatk VariantRecalibrator \
-R $reference \
-V $outdir/population/${outname}.HC.snps.VQSR.vcf.gz \
-resource:mills,known=true,training=true,truth=true,prior=12.0 $GATK_bundle/Mills_and_1000G_gold_standard.indels.hg38.vcf.gz \
-an DP -an QD -an FS -an SOR -an ReadPosRankSum -an MQRankSum\
-mode INDEL \
--max-gaussians 6 \
--rscript-file $outdir/population/${outname}.HC.indels.plots.R \
--tranches-file $outdir/population/${outname}.HC.indels.tranches \
-O $outdir/population/${outname}.HC.snps.indels.recal && \
time $gatk ApplyVQSR \
-R $reference \
-V $outdir/population/${outname}.HC.snps.VQSR.vcf.gz \
--truth-sensitivity-filter-level 99.0 \
--tranches-file $outdir/population/${outname}.HC.snps.indels.tranches \
--recal-file $outdir/population/${outname}.HC.snps.indels.recal \
-mode INDEL \
-O $outdir/population/${outname}.HC.VQSR.vcf.gz && echo "** SNPs and Indels VQSR (${sample}.HC.VQSR.vcf.gz finish) done **"
```
There are two warnings:
WARN VariantDataManager - WARNING: Training with very few variant sites! Please check the model reporting PDF to ensure the quality of the model is reliable.
WARN VariantRecalibratorEngine - Evaluate datum returned a NaN
And the program stopped due to "No data found"

I searched on this forum for solution and removed "-MQRankSum". The recalibration process passed this time although the first warning is still there:
WARN VariantDataManager - WARNING: Training with very few variant sites! Please check the model reporting PDF to ensure the quality of the model is reliable.

But the ApplyVQSR process stopped, due to:

```
13:55:06.069 INFO ApplyVQSR - Deflater: IntelDeflater
13:55:06.069 INFO ApplyVQSR - Inflater: IntelInflater
13:55:06.070 INFO ApplyVQSR - GCS max retries/reopens: 20
13:55:06.070 INFO ApplyVQSR - Requester pays: disabled
13:55:06.070 INFO ApplyVQSR - Initializing engine
13:55:06.517 INFO FeatureManager - Using codec VCFCodec to read file file:///home/pang/data/public_data/UKBB/exome_population/population/ukb_efe_chr4.HC.snps.indels.recal
13:55:06.593 INFO FeatureManager - Using codec VCFCodec to read file file:///home/pang/data/public_data/UKBB/exome_population/population/ukb_efe_chr4.HC.snps.VQSR.vcf.gz
13:55:06.776 INFO ApplyVQSR - Done initializing engine
13:55:06.778 INFO ApplyVQSR - Shutting down engine
[December 11, 2019 1:55:06 PM CET] org.broadinstitute.hellbender.tools.walkers.vqsr.ApplyVQSR done. Elapsed time: 0.02 minutes.
Runtime.totalMemory()=1933574144
***********************************************************************

A USER ERROR has occurred: Couldn't read file /home/pang/data/public_data/UKBB/exome_population/population/ukb_efe_chr4.HC.snps.indels.tranches. Error was: /home/pang/data/public_data/UKBB/exome_population/population/ukb_efe_chr4.HC.snps.indels.tranches with exception: /home/pang/data/public_data/UKBB/exome_population/population/ukb_efe_chr4.HC.snps.indels.tranches

***********************************************************************
Set the system property GATK_STACKTRACE_ON_USER_EXCEPTION (--java-options '-DGATK_STACKTRACE_ON_USER_EXCEPTION=true') to print the stack trace.
```

I do not understand what this "exception" is. Could you give me some suggestion on how to solve this?
On the other hand, I know the limitation for VQSR is above 30 exomes, in some cases it is the reason for "No data found" error. Although I did VQSR for each chromosome separately, I think 50,000 samples is enough.
Is there some method to merge these chromosomes together and do the VQSR afterwards?

Thanks!
Shichao