Calcium (Ca) sparks are the fundamental sarcoplasmic reticulum (SR) Ca release events in cardiac myocytes, and they have a typical duration of 20-40 ms. However, when a fraction of ryanodine receptors (RyRs) are blocked by tetracaine or ruthenium red, Ca sparks lasting hundreds of milliseconds have been observed experimentally. The fundamental mechanism underlying these extremely prolonged Ca sparks is not understood. In this study, we use a physiologically detailed mathematical model of subcellular Ca cycling to examine how Ca spark duration is influenced by the number of functional RyRs in a junctional cluster (which is reduced by tetracaine or ruthenium red) and other SR Ca handling properties. One RyR cluster contains a few to several hundred RyRs, and we use a four-state Markov RyR gating model. Each RyR opens stochastically and is regulated by cytosolic and luminal Ca. We varied the number of functional RyRs in the single cluster, diffusion within the SR network, diffusion between network and junctional SR, cytosolic Ca diffusion, SERCA uptake activity, and RyR open probability. For long-lasting Ca release events, opening events within the cluster must occur continuously because the typical open time of the RyR is only a few milliseconds. We found the following: 1) if the number of RyRs is too small, it is difficult to maintain consecutive openings and stochastic attrition terminates the release; 2) if the number of RyRs is too large, the depletion of Ca from the junctional SR terminates the release; and 3) very long release events require relatively small-sized RyR clusters (reducing flux as seen experimentally with tetracaine) and sufficiently rapid intra-SR Ca diffusion, such that local junctional intra-SR [Ca] can be maintained by intra-SR diffusion and overall SR Ca reuptake.