if I have time series with 1000 values , and I want to build a predictive model , how far in the future should i successfully forecast to make my predictive model valid, is there any condition or rule for this?
Time-series forecasting models are usually built with a specific objective in mind. It is that objective that should determine how the model is designed, fit, and its performance evaluated/determined to be successful.
For example, at the end of each month you may have a need to forecast next month's sales as accurately as possible. This means we are concerned about minimising our forecast error 1 month out-of-sample; we are not as concerned about the fit/R square/error of our model in-sample. We must also decide how accurate the forecast needs to be in order to determine the model a success. This is our objective.
In order to minimise forecast error 1 month out-of-sample, you might want to use a cross-validation procedure. The following link describes the process that can be used for time-series data. Rob Hyndman includes an example where forecast error is minimised 1 month out and another example where forecast error is minimised 4 months out; it merely depends on your objective.
Cross-Validation for Time-Series
So to answer your question directly, you should forecast as far into the future as your objective requires. And in order to determine whether your model is 'valid' or 'successful', you must decide what an acceptable amount of forecast error will be.