string like "1900-01-00" in func datediff,unix_timestamp will make tablet crash

[lqy222]

Bug Description

datediff and unix_timestamp func that convert string to date extract "1900-01-00" fail

Expected Behavior

Relation Case
inputs:
- columns: ["id int64", "c1 string", "c2 string", "c3 date", "c4 date", "c5 timestamp"]
indexs: ["index1: id: c5"]
rows:
- [5, "1900-01-01", "1900-01-00", "9999-12-31", "1900-01-01", 1590738991300]
sql: select id ,datediff(c1,c2) as svs,datediff(c1,c3) as svd,datediff(c3,c1) as dvs,datediff(c3,c4) as dvd from {0};

inputs:
- columns: ["id int64", "c1 string", "c2 timestamp"]
indexs: ["index1: id: c2"]
rows:
- [1, 1900-01-00, 1590738991000]
sql: select id ,unix_timestamp(c1) as unix_timestamp from {0};

---

This is a example problem of #926, null value returned by e.g timestamp(-1) become UndefValue for timestamp_to_date

[aceforeverd]

@lqy222 what's the executing environment ? online or offline. Provide the core dump trace info ?

[lqy222]

It's online env.
cli log
select c0,datediff(c1,c2) as svs,datediff(c1,c3) as svd,datediff(c3,c1) as dvs,datediff(c3,c4) as dvd from t2;

W0406 01:33:21.178077 23237 rpc_client.h:122] request error. [E1014]Got EOF of Socket{id=2 fd=13 addr=10.100.121.245:30003:47162} (0x0x671e400) [R1][E112]Not connected to 10.100.121.245:30003 yet, server_id=2 [R2][E112]Not connected to 10.100.121.245:30003 yet, server_id=2 [R3][E112]Not connected to 10.100.121.245:30003 yet, server_id=2
W0406 01:33:21.178284 23237 tablet_client.cc:94] send rpc request failed
W0406 01:33:21.178313 23237 sql_cluster_router.cc:1249] Status: [1500] Query rpc failed--[E1014]Got EOF of Socket{id=2 fd=13 addr=10.100.121.245:30003:47162} (0x0x671e400) [R1][E112]Not connected to 10.100.121.245:30003 yet, server_id=2 [R2][E112]Not connected to 10.100.121.245:30003 yet, server_id=2 [R3][E112]Not connected to 10.100.121.245:30003 yet, server_id=2--ReturnCode[0]--
Error: [1500] Query rpc failed--[E1014]Got EOF of Socket{id=2 fd=13 addr=10.100.121.245:30003:47162} (0x0x671e400) [R1][E112]Not connected to 10.100.121.245:30003 yet, server_id=2 [R2][E112]Not connected to 10.100.121.245:30003 yet, server_id=2 [R3][E112]Not connected to 10.100.121.245:30003 yet, server_id=2--ReturnCode[0]--

journalctl error log
localhost kernel: openmldb[14011]: segfault at 7fd50f865008 ip 00007fd5defe5841 sp 00007fd50f9697c0 error 4 in libgcc_s-4.8.5-20150702.so.1[7fd5defd5000+15000]

no tablet log recorded

[aceforeverd]

can't reproduce in sql engine. I'll add a feature enhance for string to timestamp/date conversion, it seems problematic for many edge cases.

[aceforeverd]

Failed cases:

• select datediff(Date(timestamp(2177481600)), Date("2021-05-01"));
• select datediff(Date(timestamp(-1)), Date("2021-05-01"));

Fixed:

• "1900-01-00" in datadiff, by feat(udf): support datediff dates before 1900 #3499