I wanted to keep things simple. To use song_ids I must to download them and save in local database or add a query asking for song_id before query with playlist. Echonest already know which song I wanted to use, so why don't use this knowledge?
BTW in description of track_id in Playlist API is written: 'Echo Nest or Rosetta IDs (See Project Rosetta Stone)'. If I use ?bucket:id:CAtalogID Echonest returning with array of CAtalogID:song:item_id, so this is Rosetta ID, but it doesn't works with track_id.