Linux service anomaly due to full inodes: Don't just check df -h during troubleshooting

Once, an online service failed to write logs. 'df -h' showed plenty of disk space left, so everyone initially thought it was a permissions issue. Later, we discovered that inodes were exhausted by small files; the disk wasn't full, but new files couldn't be created. My troubleshooting process was to first run 'df -i' to check inode usage, then use 'find' to count small files by directory…