the question is what do you want to get.

Is this just an optical image, which you want to rescale or there is a second map of something (surface emissivity, vegetation, energy flux), for which the clouds and bad pixels should be taken into account so that the maps are cleared of them? 

In any case, before making a resampling one needs to fill the gaps with valid data and here comes the variability of the parameter under consideration - for some of them, interpolation in space gives better results while for the other ones it's better to search for neighbors on the time axis.

Thank you, Dr. Artem G. Feofilov

I  have Landsat 8 30m resolution band which I would resample into 250 resolution for further analysis as I need it in that resolution.so I want to know whether the resampling would affect the result if I resample it 1st and then remove the cloud & bad pixels. 

I just want to know which should be done fast in a sequence.. resample then remove clouds or removing could then resample? which would be safe to follow.

Dear Sumit,

As Artem indicated earlier, the answer to your question depends mostly on the ultimate purpose of your investigation, as well as on the required accuracy of the information generated. From a purely logical point of view, however, it makes much more sense to first mask the clouds and cloud shadows at the high spatial resolution (30 m) of the Landsat data, than after aggregation into a low resolution product. First of all, it is much easier to detect and classify clouds and cloud shadows at high resolution. Second, the area contaminated by clouds and cloud shadows in the low resolution product will be larger than in the original Landsat data.

It is not clear why you are concerned by completing this task 'fast': the execution time largely depends on the performance of the hardware available, and somewhat on the software efficiency. In any case, new faster hardware will become available sooner or later. Except for a small number of time-critical applications, you should generally be much more concerned by the quality of your outputs instead. Another way to state this is that 'fast' is generally incompatible with 'safe' (or 'good quality').

Lastly, remember that there are various satellite-based instruments of Earth Observation that generate data at spatial resolutions around 250 m, including MODIS, MISR, MERIS, or Sentinel-3, and that those instruments provide a much more frequent coverage of your area of interest than Landsat. In other words, you may want to evaluate whether downgrading the spatial resolution of Landsat data by a factor 69 or so is really the best approach.

Cheers, Michel.