Although different graph self-supervised learning strategies have been proposed to tackle the supervision shortage issue in graph learning tasks, Graph contrastive learning (GCL) has been the most prevalent approach to this problem. Despite the remarkable performances those GCL methods have achieved, existing GCL methods that heavily depend on various manually designed augmentation techniques still struggle to improve model robustness without risking losing task-relevant information. Consequently, the learned representation is either brittle or unilluminating. In light of this, we introduce the GraphCV, which follows the information bottleneck principle to learn minimal yet sufficient representations from graph data. Specifically, our proposed model elicits the predictive (useful for downstream instance discrimination) and other non-predictive features separately. Except for the conventional contrastive loss which guarantees the consistency and sufficiency of the representations across different augmentation views, we introduce a cross-view reconstruction mechanism to pursue the disentanglement of the two learned representations. Besides, an adversarial global view is added as the third view of contrastive loss to avoid the learned representation from being drafted too far away from the original distribution. We empirically demonstrate that our proposed model outperforms the state-of-the-art on graph classification task over multiple benchmark datasets.