Probably very early, given they have the know how, but you will have to accept some degree of inconsistency between different places.
This is the the most simple design I could come think of:
To measure the seconds, you would need to invent some type of escapement.
An escapement is a mechanical linkage in mechanical watches and clocks
that provides impulses to the timekeeping element and periodically
releases the gear train to move forward, advancing the clock's hands.
The most primitive type of escapement I can come up with is based on the principle of Shishi-odoshi. As a hobby woodworker, I can say that these are trivial (and fun) to make even with rudimentary tools.
The challenge is to maintain a consistent water flow. However, this could be managed by a system of progressively smaller reservoirs that ensure the last one always maintains the same water level.
Moving from this rather fragile type of escapement to a real clock, I would suggest keeping the entire mechanism lightweight, using materials like rattan to weave the dials instead of using carved gears. The easiest approach would be to have separate but connected dials for seconds, minutes, and hours, standing next to each other. The reading would be done by poles stuck in the ground in front of the dials, while the dials are turning around their axes.
Each "clack" turns the second dial by one step, representing a "second". The motion is then transferred to the minute dial and the hour dial, and so on if needed.
The contraption itself, as you can see is not that hard to conceive, however there are other challenges:
Since we can't rely on the unique behavior of a pendulum, in order to achieve consistent measurements, you would need to normalize how fast the Shishi-odoshi is filled with water. For this, you need to have a way to communicate the amount of water the last reservoir can hold, the size of the hole through which the water is flowing out, and the dimensions and weight of the Shishi-odoshi between the individual clockmakers. Otherwise, you will end up with different second lengths across the different places.
This presumes the existence of a universal normalization of dimensions and the effective communication of specific instructions across distances, which, in turn, assumes a level of organization in society higher than what we usually imagine in prehistoric times.
Furthermore, time synchronization will be tricky. How will people in different places recognize when "Hour One" begins each day? Some societies may utilize their (usually) megalithic calendar structures to determine the moment when the sun rises on the summer solstice or something similar, but this method will always be prone to errors and not every village can be expected to construct their own Stonehenge to periodically synchronize their clocks.
In our history, we only managed to overcome this problem after developing clocks small enough to transport between places while maintaining consistent operation, and even that solution wasn't ideal.