I would say that the length of the signal does not impact the final result of your CWT. CWT analysis consists of a convolution operation between the signal itself and the mother wavelet you chose. The operation is done by translating the wavelet along the time axis.
As far as concerns the scale, if you aim to analyze some predictable pattern to emphasize it and detect it (e.g. QRS complex in ECG, foot contact in gait analysis), you better choose some appropriate wavelet type (shape similar to the signal's region you want to detect) and scale (frequency similar to that of the signal in the region of interest). You can derive such information from the visual inspection of the signal, from the fast Fourier transform of the signal in different regions, or from the literature knowledge.
Sometimes, you may want just to visualize the time-frequency representation of a signal, to get a better idea of its temporal evolution and spectral content. In this case, you can apply CWT in a wide range of frequencies, and get a 2D plot where the x-axis corresponds to the time vector and the y-axis refers to the frequency content.
I support the response given by Luigi Borzí. It is would be interesting to come with some technique to identify best suited wavelet type and scale. Most important is to have adequate number of data points to capture behaviour of the variable from short-run, mid-run, long-run to very long-run.