[LU-7050] llog_skip_over skip the record by too little minimum record size.

• Type: Bug
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.8.0
• Affects Version/s: Lustre 2.8.0
• Severity: 3
• Assignee: Di Wang
• Reporter: Di Wang

Description

During processing llog records, it needs to locate the record by the start_index, and if start_index is not zero, it will first try to skip some records to avoid reading too much useless records (see llog_process_thread). In current implementation, llog_skip_over will skip the record by LLOG_MIN_REC_SIZE, only 24 bytes, which is too less for update record (usually around 1000 bytes).

static inline void llog_skip_over(struct llog_log_hdr *llh, __u64 *off,
                                  int curr, int goal, __u32 chunk_size)
{
        if (goal > curr) {
                if (llh->llh_size == 0) {
                        /* variable size records */
                        *off = *off + (goal - curr - 1) * LLOG_MIN_REC_SIZE;
                } else {
                        *off = chunk_size + (goal - 1) * llh->llh_size;
                }
        }
        /* always align with lower chunk boundary*/
        *off &= ~(chunk_size - 1);
}              

So we probably should record the minimum record size in llog_osd_write_rec() and remember it in llh_size (in llog header), which is only used by fixed size llog record for now.

Issue Links

• is related to: LU-10027 Unable to finish mount on MDS while Rolling downgrade
• is related to: LU-6831 The ticket for tracking all DNE2 bugs