Unix sort when sequence resets

You can't easily sort some of your output one way but make an exception another way.

If only 999 and 000 were the sequences in a single given minute you could try to split the filename into fields and sort the final field (your sequence counter) in descending order. But that couldn't work because your example shows 995 and 996 also in the same minute (20200420_1755) and the same sorting criterion would apply to those too (i.e. you'd get 996 followed by 995).

So, in order to handle this specific situation we use some awk. The code extends your three digit sequence number back into a proper counting value, sorts by that, and then strips it back off again

ls node* |
    sort |                                                  # First pass attempt
    awk -F_ '
        BEGIN { OFS = FS }                                  # Delimiter always "_"
        (oseq % 1000) == 999 { thousands++ }                # After previous 999 add 1000
        { seq = $NF + (thousands * 1000) }                  # Convert sequence into value
        seq < oseq && (seq % 1000) < 500 { seq += 1000 }    # Special case
        { $5 = seq; oseq = seq; print }                     # Append counter, output result
    ' |
    sort -t_ -n -k5,6 |                                     # Sort by counter
    cut -d_ -f1-4                                           # Strip it off

Result using your example dataset

node55_20200420_1755_995
node55_20200420_1755_996
node55_20200420_1756_997
node55_20200420_1756_998
node55_20200420_1757_999
node55_20200420_1757_001
node55_20200420_1758_002
node55_20200420_1758_003

This code is not bulletproof. If you really want to generate anywhere more than 100 instances in a single minute then you shouldn't be using just a three digit sequence number.