Got the same issue with 1.2.3 and we identified the bug in this piece of code:

```

/**

• Given date, convert this instance to a regular expression.
  *
• Used to compute sub-regex when the pattern has both %d and %i, and the
• date is known.
  */
  public String toRegexForFixedDate(Date date) {
  StringBuilder buf = new StringBuilder();
  Converter<Object> p = headTokenConverter;
  while (p != null) {
  if (p instanceof LiteralConverter) {
  buf.append(p.convert(null));
  } else if (p instanceof IntegerTokenConverter) {
  buf.append("(
  d{1,3})");
  } else if (p instanceof DateTokenConverter) {
  buf.append(p.convert(date));
  }
  p = p.getNext();
  }
  return buf.toString();
  }
  ```

and the problem happens when we have more than 1000 files created. Indeed, the above method only looks for file with index made of up to 3 digits (see `\d{1,3}`)  while it should look for files with at least one digit (i.e. `\d+`).

So basically it will keep `totalSizeCap/maxFileSize` files + all files with index >= 1000.

And basically, after checking head of code we noticed it has a fix and it is described in https://jira.qos.ch/browse/LOGBACK-1175