raid 5 mdadm replace with read errors

I have a software RAID 5 with 3 disks. For a few days, sda has been giving errors, and today the new disk arrived. I launched mdadm --replace and the process started correctly, but due to read errors on the disk, the operation is taking a very long time (over a week for about 1TB). Should I be patient, or can I use --remove sda? And force the recovery using parity information instead? What do you recommend?