Add support for UUID version 7

Although the specification for UUIDv7 is still in draft, the UUIDv7 algorithm has been relatively stable as it progresses to completion. Version 7 UUIDs can be very useful, because they are lexographically sortable, which can improve e.g: database index locality. See section 6.10 of the draft specification for further explanation: https://datatracker.ietf.org/doc/draft-ietf-uuidrev-rfc4122bis/ The specification allows up to 12 bits of extra timestamp precision, to make UUID generation closer to monotonically increasing. This provides between 1ms and ~240ns of timestamp precision. At the cost of some code complexity and a small performance penalty, a kwarg may specify any arbitrary precision between 0 and 12 extra bits. Any stronger guarantees of monotonicity have considerably larger tradeoffs, so nothing more is implemented. This limitation is documented. Ruby issue: https://bugs.ruby-lang.org/issues/19735

Author: nevans

lib/random/formatter.rb

@@ -174,6 +174,125 @@ def uuid
  "%08x-%04x-%04x-%04x-%04x%08x" % ary
  end
 
+ alias uuid_v4 uuid
+
+ # Generate a random v7 UUID (Universally Unique IDentifier).
+ #
+ # require 'random/formatter'
+ #
+ # Random.uuid_v7 # => "0188d4c3-1311-7f96-85c7-242a7aa58f1e"
+ # Random.uuid_v7 # => "0188d4c3-16fe-744f-86af-38fa04c62bb5"
+ # Random.uuid_v7 # => "0188d4c3-1af8-764f-b049-c204ce0afa23"
+ # Random.uuid_v7 # => "0188d4c3-1e74-7085-b14f-ef6415dc6f31"
+ # # |<--sorted-->| |<----- random ---->|
+ #
+ # # or
+ # prng = Random.new
+ # prng.uuid_v7 # => "0188ca51-5e72-7950-a11d-def7ff977c98"
+ #
+ # The version 7 UUID starts with the least significant 48 bits of a 64 bit
+ # Unix timestamp (milliseconds since the epoch) and fills the remaining bits
+ # with random data, excluding the version and variant bits.
+ #
+ # This allows version 7 UUIDs to be sorted by creation time. Time ordered
+ # UUIDs can be used for better database index locality of newly inserted
+ # records, which may have a significant performance benefit compared to random
+ # data inserts.
+ #
+ # The result contains 74 random bits (9.25 random bytes).
+ #
+ # Note that this method cannot be made reproducable with Kernel#srand, which
+ # can only affect the random bits. The sorted bits will still be based on
+ # Process.clock_gettime.
+ #
+ # See draft-ietf-uuidrev-rfc4122bis[https://datatracker.ietf.org/doc/draft-ietf-uuidrev-rfc4122bis/]
+ # for details of UUIDv7.
+ #
+ # ==== Monotonicity
+ #
+ # UUIDv7 has millisecond precision by default, so multiple UUIDs created
+ # within the same millisecond are not issued in monotonically increasing
+ # order. To create UUIDs that are time-ordered with sub-millisecond
+ # precision, up to 12 bits of additional timestamp may added with
+ # +extra_timestamp_bits+. The extra timestamp precision comes at the expense
+ # of random bits. Setting <tt>extra_timestamp_bits: 12</tt> provides ~244ns
+ # of precision, but only 62 random bits (7.75 random bytes).
+ #
+ # prng = Random.new
+ # Array.new(4) { prng.uuid_v7(extra_timestamp_bits: 12) }
+ # # =>
+ # ["0188d4c7-13da-74f9-8b53-22a786ffdd5a",
+ # "0188d4c7-13da-753b-83a5-7fb9b2afaeea",
+ # "0188d4c7-13da-754a-88ea-ac0baeedd8db",
+ # "0188d4c7-13da-7557-83e1-7cad9cda0d8d"]
+ # # |<--- sorted --->| |<-- random --->|
+ #
+ # Array.new(4) { prng.uuid_v7(extra_timestamp_bits: 8) }
+ # # =>
+ # ["0188d4c7-3333-7a95-850a-de6edb858f7e",
+ # "0188d4c7-3333-7ae8-842e-bc3a8b7d0cf9", # <- out of order
+ # "0188d4c7-3333-7ae2-995a-9f135dc44ead", # <- out of order
+ # "0188d4c7-3333-7af9-87c3-8f612edac82e"]
+ # # |<--- sorted -->||<---- random --->|
+ #
+ # Any rollbacks of the system clock will break monotonicity. UUIDv7 is based
+ # on UTC, which excludes leap seconds and can rollback the clock. To avoid
+ # this, the system clock can synchronize with an NTP server configured to use
+ # a "leap smear" approach. NTP or PTP will also be needed to synchronize
+ # across distributed nodes.
+ #
+ # Counters and other mechanisms for stronger guarantees of monotonicity are
+ # not implemented. Applications with stricter requirements should follow
+ # {Section 6.2}[https://www.ietf.org/archive/id/draft-ietf-uuidrev-rfc4122bis-07.html#monotonicity_counters]
+ # of the specification.
+ #
+ def uuid_v7(extra_timestamp_bits: 0)
+ case (extra_timestamp_bits = Integer(extra_timestamp_bits))
+ when 0 # min timestamp precision
+ ms = Process.clock_gettime(Process::CLOCK_REALTIME, :millisecond)
+ rand = random_bytes(10)
+ rand.setbyte(0, rand.getbyte(0) & 0x0f | 0x70) # version
+ rand.setbyte(2, rand.getbyte(2) & 0x3f | 0x80) # variant
+ "%08x-%04x-%s" % [
+ (ms & 0x0000_ffff_ffff_0000) >> 16,
+ (ms & 0x0000_0000_0000_ffff),
+ rand.unpack("H4H4H12").join("-")
+ ]
+
+ when 12 # max timestamp precision
+ ms, ns = Process.clock_gettime(Process::CLOCK_REALTIME, :nanosecond)
+ .divmod(1_000_000)
+ extra_bits = ns * 4096 / 1_000_000
+ rand = random_bytes(8)
+ rand.setbyte(0, rand.getbyte(0) & 0x3f | 0x80) # variant
+ "%08x-%04x-7%03x-%s" % [
+ (ms & 0x0000_ffff_ffff_0000) >> 16,
+ (ms & 0x0000_0000_0000_ffff),
+ extra_bits,
+ rand.unpack("H4H12").join("-")
+ ]
+
+ when (0..12) # the generic version is slower than the special cases above
+ rand_a, rand_b1, rand_b2, rand_b3 = random_bytes(10).unpack("nnnN")
+ rand_mask_bits = 12 - extra_timestamp_bits
+ ms, ns = Process.clock_gettime(Process::CLOCK_REALTIME, :nanosecond)
+ .divmod(1_000_000)
+ "%08x-%04x-%04x-%04x-%04x%08x" % [
+ (ms & 0x0000_ffff_ffff_0000) >> 16,
+ (ms & 0x0000_0000_0000_ffff),
+ 0x7000 |
+ ((ns * (1 << extra_timestamp_bits) / 1_000_000) << rand_mask_bits) |
+ rand_a & ((1 << rand_mask_bits) - 1),
+ 0x8000 | (rand_b1 & 0x3fff),
+ rand_b2,
+ rand_b3
+ ]
+
+ else
+ raise ArgumentError, "extra_timestamp_bits must be in 0..12"
+ end
+ end
+
  private def gen_random(n)
  self.bytes(n)
  end

test/ruby/test_random_formatter.rb

@@ -75,6 +75,54 @@ def test_uuid
  assert_match(/\A\h{8}-\h{4}-\h{4}-\h{4}-\h{12}\z/, uuid)
  end
 
+ def test_uuid_v7(extra_timestamp_bits)
+ t1 = current_uuid7_time
+ uuid = @it.uuid_v7
+ t3 = current_uuid7_time
+
+ assert_match(/\A\h{8}-\h{4}-7\h{3}-[89ab]\h{3}-\h{12}\z/, uuid)
+
+ t2 = get_uuid7_time(uuid)
+ assert_operator(t1, :<=, t2)
+ assert_operator(t2, :<=, t3)
+ end
+
+ def test_uuid_v7_extra_timestamp_bits
+ 0.upto(12) do |extra_timestamp_bits|
+ t1 = current_uuid7_time extra_timestamp_bits: extra_timestamp_bits
+ uuid = @it.uuid_v7 extra_timestamp_bits: extra_timestamp_bits
+ t3 = current_uuid7_time extra_timestamp_bits: extra_timestamp_bits
+
+ assert_match(/\A\h{8}-\h{4}-7\h{3}-[89ab]\h{3}-\h{12}\z/, uuid)
+
+ t2 = get_uuid7_time uuid, extra_timestamp_bits: extra_timestamp_bits
+ assert_operator(t1, :<=, t2)
+ assert_operator(t2, :<=, t3)
+ end
+ end
+
+ # It would be nice to simply use Time#floor here. But that is problematic
+ # due to the difference between decimal vs binary fractions.
+ def current_uuid7_time(extra_timestamp_bits: 0)
+ denominator = (1 << extra_timestamp_bits).to_r
+ Process.clock_gettime(Process::CLOCK_REALTIME, :nanosecond)
+ .then {|ns| ((ns / 1_000_000r) * denominator).floor / denominator }
+ .then {|ms| Time.at(ms / 1000r, in: "+00:00") }
+ end
+
+ def get_uuid7_time(uuid, extra_timestamp_bits: 0)
+ denominator = (1 << extra_timestamp_bits) * 1000r
+ extra_chars = extra_timestamp_bits / 4
+ last_char_bits = extra_timestamp_bits % 4
+ extra_chars += 1 if last_char_bits != 0
+ timestamp_re = /\A(\h{8})-(\h{4})-7(\h{#{extra_chars}})/
+ timestamp_chars = uuid.match(timestamp_re).captures.join
+ timestamp = timestamp_chars.to_i(16)
+ timestamp >>= 4 - last_char_bits unless last_char_bits == 0
+ timestamp /= denominator
+ Time.at timestamp, in: "+00:00"
+ end
+
  def test_alphanumeric
  65.times do |n|
  an = @it.alphanumeric(n)